diff --git a/.github/workflows/jekyll.yml b/.github/workflows/jekyll.yml index c6522c5..dc13bb7 100644 --- a/.github/workflows/jekyll.yml +++ b/.github/workflows/jekyll.yml @@ -53,6 +53,10 @@ jobs: # Before the build: these plugins decide every page's description, order # and read time. run: bundle exec ruby test/run_all.rb + - name: Unit-test the search script + # js/search.js's matching and ranking, in node's own test runner: no + # packages to install. The runner image ships Node. + run: node --test 'test/js/*.test.js' - name: Lint the post sources # Front matter against _data/post_conventions.yml, plus table, math and # heading rules the built HTML no longer shows. diff --git a/_data/post_conventions.yml b/_data/post_conventions.yml index a6d860f..f0a0151 100644 --- a/_data/post_conventions.yml +++ b/_data/post_conventions.yml @@ -38,6 +38,10 @@ topics: # (_plugins/posted_date.rb). forbidden_keys: [author, last_modified_at] +# A title longer than this needs a shorter `seo_title` for
+
+
Of the three options, start with the one requiring the least work: managed. [Amazon Bedrock Knowledge Bases GraphRAG](https://aws.amazon.com/blogs/machine-learning/announcing-general-availability-of-amazon-bedrock-knowledge-bases-graphrag-with-amazon-neptune-analytics/) reached general availability in March 2025 and runs on Amazon Neptune Analytics. Drop documents in and it generates embeddings and an entity-relationship graph automatically, no graph modeling expertise required. The [AWS ML blog's build guide](https://aws.amazon.com/blogs/machine-learning/build-graphrag-applications-using-amazon-bedrock-knowledge-bases/) unpacks the internals: an `ExtractChunkEntity` step uses an LLM to pull entities from each chunk and stores them as **three node types** ("The system creates three types of nodes: chunk, document, and entity"), and at query time it finds top-k chunks by vector search and then traverses their neighboring graph. The post states they chose Titan Text Embeddings v2 as the embedding model.
@@ -315,8 +316,8 @@ One [independent measurement](https://arxiv.org/abs/2502.11371) found constructi
The first prescription came from KET-RAG itself, the same paper that produced the $33,000 estimate, and it's captured in their indexing diagram. Instead of sweeping the whole corpus with an LLM, extract only the key chunks with conventional GraphRAG to build a **KG skeleton**, and lay down a text-keyword bipartite graph over the rest using word tokenization with no LLM at all — a two-tier structure. At query time, a ratio $$\theta$$ splits how much gets retrieved from each tier: the expensive tier thin, the cheap tier thick.
-
-
+
+
How thick the cheap tier can get is ultimately an empirical question, but the direction is clear: spreading the LLM uniformly across all of indexing was waste. The second prescription pushes the same observation to its limit — not using the LLM sparingly in indexing but removing it entirely. That's Microsoft's [**LazyGraphRAG**](https://www.microsoft.com/en-us/research/blog/lazygraphrag-setting-a-new-standard-for-quality-and-cost/) from November 2024. The inversion: don't use an LLM at indexing time. Capture structure cheaply with NLP noun phrase extraction and **defer LLM use to query time.** The numbers Microsoft published:
@@ -343,8 +344,8 @@ The third prescription comes from academic follow-ups. Where the first two touch
What original GraphRAG and the two follow-ups just described pick up from the same graph is compressed into one figure from the PathRAG paper. Red marks what actually enters the prompt.
-
-
+
+
Whole communities in (a), whole neighborhoods in (b), only paths in (c) — moving left to right, the tokens entering the prompt shrink, and this lineage's finding is that the shrunken side answers better. Once you've laid down a graph, the next fight isn't "what should we put in" but **"what can we leave out."** That said, cost coming down doesn't mean quality follows, and a bill that shifted from upfront to metered isn't always favorable — §7.5 revisits both calculations with independent numbers.
@@ -619,8 +620,8 @@ More uncomfortable is the variance between models. Same corpus, same procedure,
The paper patches this fragmentation with post-processing: splitting compound nodes that swallowed whole sentences, then grouping scattered fragments under higher-level concepts. How one sentence changes through that process is captured in the paper's first figure. Take a sentence reporting that TV viewing at 12 months is negatively associated with cognitive ability at 4.5 years. Initial extraction produces one compound node lumping the whole sentence together (association between … and …). Splitting decomposes it into individual entities, and abstraction groups them under higher concepts like infant media exposure to connect the scattered islands. The color coding marks what entity deduplication, pronoun resolution, and provenance tracking each contribute at every stage.
-
-
+
+
Here are the numbers after all three stages. Post-processing lifts every model's $$F_{GC}$$. But the gaps between models remain nearly intact. Initial extraction's 0.249–0.794 merely shifts to 0.441–0.927.
@@ -660,8 +661,8 @@ The paper prescribes two remedies: route by query type, or integrate both result
These outcomes mostly survive adding retrieval post-processing. The same evaluation crossed in two techniques. **Reranking** rescores retrieved chunks to change their order. **IRCoT** alternates retrieval and reasoning, asking after each round what else to look for given the evidence gathered so far, rather than answering in one shot. Both were applied, and the per-query-type advantages held.
-
-
+
+
This figure shows one thing: the ordering flips between the two panels. On single-hop (NQ), vector RAG stays on top across all three strategies; on multi-hop (MultiHop-RAG), graph-based HippoRAG2 stays on top across all three. Post-processing can't change the ordering among methods. It merely lifts everyone's scores together, which means it won't substitute for choosing a method.
diff --git a/_posts/2026-07-27-ai-ready-data-semantic-layer-knowledge-graph.md b/_posts/2026-07-27-ai-ready-data-semantic-layer-knowledge-graph.md
index 5ab896b..4bbe937 100644
--- a/_posts/2026-07-27-ai-ready-data-semantic-layer-knowledge-graph.md
+++ b/_posts/2026-07-27-ai-ready-data-semantic-layer-knowledge-graph.md
@@ -1,6 +1,7 @@
---
layout: post
title: "데이터에 맥락을 새겨 넣는 법 — AI Ready Data, Semantic Layer, Knowledge Graph, Ontology"
+seo_title: "데이터에 맥락을 새겨 넣는 법 — 시맨틱 레이어와 지식 그래프"
description: >-
'AI-ready 데이터'는 깨끗한 데이터가 아니라 맥락을 기계가 실행할 수 있는 형태로 새겨 넣은
데이터입니다. 같은 스키마에서 text-to-SQL 정확도가 16.7%에서 54.2%로 오른 이유를 짚습니다.
@@ -271,8 +272,8 @@ LLM으로 그래프를 자동 구축하는 파이프라인은 다섯 단계로
같은 뼈대를 학계 쪽 어휘로 그린 지도가 Peng et al. 서베이의 그림입니다. 인덱싱·검색·생성 사이에 축이 하나 더 놓여 있다는 게 요점입니다 — 검색 결과를 노드·트리플·경로·부분그래프 중 무엇으로 뽑을 것인가. §4.4에서 볼 LightRAG·PathRAG의 차이가 정확히 이 축에서 갈립니다.
-
-
+
+
@@ -317,8 +318,8 @@ AWS 블로그에 실린 구성도가 이 관리형의 경계를 그대로 보여
첫째 처방은 $33,000을 추정한 KET-RAG 연구진이 내놨고, 그 구조가 논문의 인덱싱 그림에 담겨 있습니다. 코퍼스 전체를 LLM으로 훑는 대신 핵심 조각들만 기존 GraphRAG로 뽑아 **KG Skeleton**(골격 그래프)을 만듭니다. 나머지는 LLM 없이 단어를 토큰화해 텍스트-키워드 이분 그래프로 구성합니다. 질의 시점에는 두 층에서 각각 얼마씩 회수할지를 비율 $$\theta$$로 나눠 씁니다. 비싼 층은 얇게, 값싼 층은 두껍게 긁는 셈입니다.
-
-
+
+
값싼 층을 어디까지 두껍게 해도 되는지는 결국 실측 문제지만, 방향은 분명합니다. LLM을 인덱싱 전 구간에 균일하게 뿌린 것이 낭비였습니다. 둘째 처방은 같은 관찰을 끝까지 밀어붙입니다. 인덱싱에서 LLM을 적게 쓰는 데 그치지 않고 아예 뺍니다. 2024년 11월 Microsoft의 [**LazyGraphRAG**](https://www.microsoft.com/en-us/research/blog/lazygraphrag-setting-a-new-standard-for-quality-and-cost/)입니다. 발상의 전환은 이렇습니다 — 인덱싱 시점에 LLM을 쓰지 말고, NLP 명사구 추출로 값싸게 구조만 잡은 뒤 **LLM 사용을 쿼리 시점으로 미룬다.** Microsoft가 공개한 결과는 다음과 같습니다.
@@ -345,8 +346,8 @@ LazyGraphRAG의 발상을 조금 더 풀어 보면 왜 이게 통하는지 보
원조 GraphRAG와 방금 본 두 후속(LightRAG·PathRAG)이 같은 그래프에서 무엇을 집어 오는지를 PathRAG 논문의 그림 하나가 압축해 보여 줍니다. 빨간색이 프롬프트에 실제로 들어가는 부분입니다.
-
-
+
+
(a)는 커뮤니티 전체를, (b)는 이웃 전체를, (c)는 경로만 사용합니다. 왼쪽에서 오른쪽으로 갈수록 프롬프트에 들어가는 토큰이 줄어듭니다. 두 연구 모두 토큰을 적게 넣은 쪽이 답을 더 잘 맞혔다고 보고합니다. 그래프를 놓고 나면 다음 싸움은 "무엇을 넣을까"가 아니라 **"무엇을 빼도 되는가"**가 됩니다. 다만 비용이 내렸다는 게 품질까지 따라온다는 뜻은 아니고, 선불에서 종량으로 바뀐 청구서가 늘 유리하다는 뜻도 아닙니다 — 두 셈 모두 §7.5에서 독립 평가 수치로 다시 따집니다.
@@ -624,8 +625,8 @@ $$\mathrm{score}(d) = \sum_{r} \frac{w_r}{60 + \mathrm{rank}_r(d)}$$
논문은 이 흩어짐을 후처리로 메웁니다. 문장을 통째로 삼킨 덩어리(compound) 노드를 쪼개고, 흩어진 조각들을 상위 개념으로 묶는 두 단계입니다. 한 문장이 그 과정을 거치며 어떻게 달라지는지가 논문의 첫 그림에 그대로 담겨 있습니다. 생후 12개월 TV 시청이 4.5세 인지능력과 음의 상관을 보인다는 문장을 놓고, 초기 추출은 문장 전체를 뭉친 덩어리 노드(association between … and …) 하나를 만들어 버립니다. 쪼개기(splitting)가 그걸 개별 엔티티로 분해하고, 추상화(abstraction)가 유아 미디어 노출 같은 상위 개념으로 묶어 흩어진 섬들을 잇습니다. 색으로 표시된 엔티티 중복 제거·대명사 해소·출처 추적이 각 단계에서 하는 일입니다.
-
-
+
+
세 단계를 다 거친 뒤의 숫자는 이렇습니다. 후처리는 모든 모델의 $$F_{GC}$$를 끌어올립니다. 그런데 모델 사이의 격차는 거의 그대로 남습니다. 초기 추출의 0.249~0.794가 후처리 후 0.441~0.927로 옮겨 갈 뿐입니다.
@@ -665,8 +666,8 @@ $$\mathrm{score}(d) = \sum_{r} \frac{w_r}{60 + \mathrm{rank}_r(d)}$$
이 승패는 검색 후처리를 얹어도 대체로 뒤집히지 않습니다. 같은 평가가 후처리 두 가지까지 교차해 봤습니다. **재순위화(reranking)**는 회수한 조각을 다시 점수 매겨 순서를 바꿉니다. **IRCoT**는 한 번에 답하지 않고, 지금까지 모은 근거로 다음에 무엇을 더 찾아야 하는지를 되물으며 검색과 추론을 번갈아 돌립니다. 둘 다 얹어 봤지만 질의 유형별 우위 관계는 그대로였습니다.
-
-
+
+
이 그림이 보여 주는 건 하나입니다. 두 패널의 순서가 서로 뒤집혀 있습니다. 단일 홉(NQ)에서는 벡터 RAG가 세 전략 내내 맨 위에 있고, 멀티홉(MultiHop-RAG)에서는 그래프를 쓰는 HippoRAG2가 세 전략 내내 맨 위에 있습니다. 후처리는 방법 사이의 순서를 바꾸지 못하고 각자의 점수를 함께 밀어 올리는 데 그칩니다. 방법 선택을 대신해 주지는 못한다는 뜻입니다.
diff --git a/assets/images/ai-ready-data-semantic-layer-knowledge-graph-1.png b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-1.png
new file mode 100644
index 0000000..753119c
Binary files /dev/null and b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-1.png differ
diff --git a/assets/images/ai-ready-data-semantic-layer-knowledge-graph-2.png b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-2.png
new file mode 100644
index 0000000..851d2ed
Binary files /dev/null and b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-2.png differ
diff --git a/assets/images/ai-ready-data-semantic-layer-knowledge-graph-3.png b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-3.png
new file mode 100644
index 0000000..66c0b31
Binary files /dev/null and b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-3.png differ
diff --git a/assets/images/ai-ready-data-semantic-layer-knowledge-graph-4.png b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-4.png
new file mode 100644
index 0000000..e864ded
Binary files /dev/null and b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-4.png differ
diff --git a/assets/images/ai-ready-data-semantic-layer-knowledge-graph-5.png b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-5.png
new file mode 100644
index 0000000..a7d9a2f
Binary files /dev/null and b/assets/images/ai-ready-data-semantic-layer-knowledge-graph-5.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-1.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-1.png
new file mode 100644
index 0000000..6eb133a
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-1.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-2.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-2.png
new file mode 100644
index 0000000..8fc7a97
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-2.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-3.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-3.png
new file mode 100644
index 0000000..d4983a1
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-3.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-4.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-4.png
new file mode 100644
index 0000000..55d8771
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-4.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-5.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-5.png
new file mode 100644
index 0000000..b3d2e31
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-5.png differ
diff --git a/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-6.png b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-6.png
new file mode 100644
index 0000000..6ac546c
Binary files /dev/null and b/assets/images/deepseek-coder-when-the-large-language-model-meets-programming-the-rise-of-code-intelligence-6.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-1.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-1.png
new file mode 100644
index 0000000..cdb2852
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-1.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-2.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-2.png
new file mode 100644
index 0000000..b64711c
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-2.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-3.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-3.png
new file mode 100644
index 0000000..7007bbd
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-3.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-4.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-4.png
new file mode 100644
index 0000000..3b33b0f
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-4.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-5.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-5.png
new file mode 100644
index 0000000..3bb88d5
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-5.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-6.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-6.png
new file mode 100644
index 0000000..709b0cc
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-6.png differ
diff --git a/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-7.png b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-7.png
new file mode 100644
index 0000000..4e3e377
Binary files /dev/null and b/assets/images/deepseek-llm-scaling-open-source-language-models-with-longtermism-7.png differ
diff --git a/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-1.png b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-1.png
new file mode 100644
index 0000000..1a88856
Binary files /dev/null and b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-1.png differ
diff --git a/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-2.png b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-2.png
new file mode 100644
index 0000000..964a7a4
Binary files /dev/null and b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-2.png differ
diff --git a/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-3.png b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-3.png
new file mode 100644
index 0000000..0f23557
Binary files /dev/null and b/assets/images/deepseek-r1--incentivizing-reasoning-capability-in-llms-via-reinforcement-learning-3.png differ
diff --git a/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-1.png b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-1.png
new file mode 100644
index 0000000..c16716b
Binary files /dev/null and b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-1.png differ
diff --git a/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-2.png b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-2.png
new file mode 100644
index 0000000..56f59db
Binary files /dev/null and b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-2.png differ
diff --git a/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-3.png b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-3.png
new file mode 100644
index 0000000..bc64486
Binary files /dev/null and b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-3.png differ
diff --git a/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-4.png b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-4.png
new file mode 100644
index 0000000..e92b31c
Binary files /dev/null and b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-4.png differ
diff --git a/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-5.png b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-5.png
new file mode 100644
index 0000000..1127c4b
Binary files /dev/null and b/assets/images/deepseek-v2--a-strong--economical--and-efficient-mixture-of-experts-language-model-5.png differ
diff --git a/assets/images/deepseek-v3-technical-report-1.png b/assets/images/deepseek-v3-technical-report-1.png
new file mode 100644
index 0000000..87f2f99
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-1.png differ
diff --git a/assets/images/deepseek-v3-technical-report-10.png b/assets/images/deepseek-v3-technical-report-10.png
new file mode 100644
index 0000000..6ceef87
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-10.png differ
diff --git a/assets/images/deepseek-v3-technical-report-11.png b/assets/images/deepseek-v3-technical-report-11.png
new file mode 100644
index 0000000..e6eb8ac
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-11.png differ
diff --git a/assets/images/deepseek-v3-technical-report-2.png b/assets/images/deepseek-v3-technical-report-2.png
new file mode 100644
index 0000000..2aa8894
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-2.png differ
diff --git a/assets/images/deepseek-v3-technical-report-3.png b/assets/images/deepseek-v3-technical-report-3.png
new file mode 100644
index 0000000..4477671
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-3.png differ
diff --git a/assets/images/deepseek-v3-technical-report-4.png b/assets/images/deepseek-v3-technical-report-4.png
new file mode 100644
index 0000000..ee73fdd
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-4.png differ
diff --git a/assets/images/deepseek-v3-technical-report-5.png b/assets/images/deepseek-v3-technical-report-5.png
new file mode 100644
index 0000000..3db496e
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-5.png differ
diff --git a/assets/images/deepseek-v3-technical-report-6.png b/assets/images/deepseek-v3-technical-report-6.png
new file mode 100644
index 0000000..68ae053
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-6.png differ
diff --git a/assets/images/deepseek-v3-technical-report-7.png b/assets/images/deepseek-v3-technical-report-7.png
new file mode 100644
index 0000000..a3d0d42
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-7.png differ
diff --git a/assets/images/deepseek-v3-technical-report-8.png b/assets/images/deepseek-v3-technical-report-8.png
new file mode 100644
index 0000000..9be76eb
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-8.png differ
diff --git a/assets/images/deepseek-v3-technical-report-9.png b/assets/images/deepseek-v3-technical-report-9.png
new file mode 100644
index 0000000..2cf21cc
Binary files /dev/null and b/assets/images/deepseek-v3-technical-report-9.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-1.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-1.png
new file mode 100644
index 0000000..4d051f8
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-1.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-2.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-2.png
new file mode 100644
index 0000000..42183de
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-2.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-3.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-3.png
new file mode 100644
index 0000000..02a8b6e
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-3.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-4.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-4.png
new file mode 100644
index 0000000..f434b17
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-4.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-5.jpg b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-5.jpg
new file mode 100644
index 0000000..17a52cc
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-5.jpg differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-6.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-6.png
new file mode 100644
index 0000000..325b345
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-6.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-7.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-7.png
new file mode 100644
index 0000000..aeef98a
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-7.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-8.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-8.png
new file mode 100644
index 0000000..bd6feea
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-8.png differ
diff --git a/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-9.png b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-9.png
new file mode 100644
index 0000000..bc2d0b5
Binary files /dev/null and b/assets/images/deepseek-v3.2--pushing-the-frontier-of-open-large-language-models-9.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-1.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-1.png
new file mode 100644
index 0000000..8ec410a
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-1.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-2.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-2.png
new file mode 100644
index 0000000..eef45b6
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-2.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-3.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-3.png
new file mode 100644
index 0000000..8433477
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-3.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-4.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-4.png
new file mode 100644
index 0000000..2efcfe7
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-4.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-5.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-5.png
new file mode 100644
index 0000000..ff1de46
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-5.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-6.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-6.png
new file mode 100644
index 0000000..ae73568
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-6.png differ
diff --git a/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-7.png b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-7.png
new file mode 100644
index 0000000..1fe22f2
Binary files /dev/null and b/assets/images/deepseekmath-pushing-the-limits-of-mathematical-reasoning-in-open-language-models-7.png differ
diff --git a/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-1.png b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-1.png
new file mode 100644
index 0000000..8e3e34d
Binary files /dev/null and b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-1.png differ
diff --git a/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-2.png b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-2.png
new file mode 100644
index 0000000..7fdb9ba
Binary files /dev/null and b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-2.png differ
diff --git a/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-3.png b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-3.png
new file mode 100644
index 0000000..199e40f
Binary files /dev/null and b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-3.png differ
diff --git a/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-4.png b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-4.png
new file mode 100644
index 0000000..07b1f08
Binary files /dev/null and b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-4.png differ
diff --git a/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-5.png b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-5.png
new file mode 100644
index 0000000..220d7b4
Binary files /dev/null and b/assets/images/deepseekmoe-towards-ultimate-expert-specialization-in-mixture-of-experts-language-models-5.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-1.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-1.png
new file mode 100644
index 0000000..c83a0a5
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-1.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-2.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-2.png
new file mode 100644
index 0000000..d78acd4
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-2.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-3.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-3.png
new file mode 100644
index 0000000..58dd529
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-3.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-4.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-4.png
new file mode 100644
index 0000000..cfcee7f
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-4.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-5.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-5.png
new file mode 100644
index 0000000..0590d66
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-5.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-6.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-6.png
new file mode 100644
index 0000000..dd461d7
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-6.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-7.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-7.png
new file mode 100644
index 0000000..887ff09
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-7.png differ
diff --git a/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-8.png b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-8.png
new file mode 100644
index 0000000..b5398a6
Binary files /dev/null and b/assets/images/direct-preference-optimization--your-language-model-is-secretly-a-reward-model-8.png differ
diff --git a/assets/images/gemma-2--improving-open-language-models-at-a-practical-size-1.png b/assets/images/gemma-2--improving-open-language-models-at-a-practical-size-1.png
new file mode 100644
index 0000000..03b7292
Binary files /dev/null and b/assets/images/gemma-2--improving-open-language-models-at-a-practical-size-1.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-1.png b/assets/images/kimi-k2--open-agentic-intelligence-1.png
new file mode 100644
index 0000000..4f91652
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-1.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-10.png b/assets/images/kimi-k2--open-agentic-intelligence-10.png
new file mode 100644
index 0000000..8e6c30c
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-10.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-11.png b/assets/images/kimi-k2--open-agentic-intelligence-11.png
new file mode 100644
index 0000000..ed8a864
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-11.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-12.png b/assets/images/kimi-k2--open-agentic-intelligence-12.png
new file mode 100644
index 0000000..d33f821
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-12.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-13.png b/assets/images/kimi-k2--open-agentic-intelligence-13.png
new file mode 100644
index 0000000..9967246
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-13.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-14.png b/assets/images/kimi-k2--open-agentic-intelligence-14.png
new file mode 100644
index 0000000..11a702e
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-14.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-15.png b/assets/images/kimi-k2--open-agentic-intelligence-15.png
new file mode 100644
index 0000000..2309ccc
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-15.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-16.png b/assets/images/kimi-k2--open-agentic-intelligence-16.png
new file mode 100644
index 0000000..db5edd5
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-16.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-2.png b/assets/images/kimi-k2--open-agentic-intelligence-2.png
new file mode 100644
index 0000000..5fb4699
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-2.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-3.png b/assets/images/kimi-k2--open-agentic-intelligence-3.png
new file mode 100644
index 0000000..fba5774
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-3.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-4.png b/assets/images/kimi-k2--open-agentic-intelligence-4.png
new file mode 100644
index 0000000..65fe5a0
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-4.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-5.png b/assets/images/kimi-k2--open-agentic-intelligence-5.png
new file mode 100644
index 0000000..4670329
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-5.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-6.png b/assets/images/kimi-k2--open-agentic-intelligence-6.png
new file mode 100644
index 0000000..c4d9f52
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-6.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-7.png b/assets/images/kimi-k2--open-agentic-intelligence-7.png
new file mode 100644
index 0000000..e070c47
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-7.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-8.png b/assets/images/kimi-k2--open-agentic-intelligence-8.png
new file mode 100644
index 0000000..ba943a9
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-8.png differ
diff --git a/assets/images/kimi-k2--open-agentic-intelligence-9.png b/assets/images/kimi-k2--open-agentic-intelligence-9.png
new file mode 100644
index 0000000..9e8425a
Binary files /dev/null and b/assets/images/kimi-k2--open-agentic-intelligence-9.png differ
diff --git a/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-1.png b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-1.png
new file mode 100644
index 0000000..7e02051
Binary files /dev/null and b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-1.png differ
diff --git a/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-2.png b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-2.png
new file mode 100644
index 0000000..da58af2
Binary files /dev/null and b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-2.png differ
diff --git a/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-3.png b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-3.png
new file mode 100644
index 0000000..7e02051
Binary files /dev/null and b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-3.png differ
diff --git a/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-4.png b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-4.png
new file mode 100644
index 0000000..9b8183a
Binary files /dev/null and b/assets/images/llama-2-open-foundation-and-fine-tuned-chat-models-4.png differ
diff --git a/assets/images/mistral-7b-1.png b/assets/images/mistral-7b-1.png
new file mode 100644
index 0000000..1f57691
Binary files /dev/null and b/assets/images/mistral-7b-1.png differ
diff --git a/assets/images/mistral-7b-2.png b/assets/images/mistral-7b-2.png
new file mode 100644
index 0000000..be63206
Binary files /dev/null and b/assets/images/mistral-7b-2.png differ
diff --git a/assets/images/mistral-7b-3.png b/assets/images/mistral-7b-3.png
new file mode 100644
index 0000000..fe1df3d
Binary files /dev/null and b/assets/images/mistral-7b-3.png differ
diff --git a/assets/images/mixtral-of-experts-1.png b/assets/images/mixtral-of-experts-1.png
new file mode 100644
index 0000000..8f0ebf6
Binary files /dev/null and b/assets/images/mixtral-of-experts-1.png differ
diff --git a/assets/images/mixtral-of-experts-2.png b/assets/images/mixtral-of-experts-2.png
new file mode 100644
index 0000000..f13b438
Binary files /dev/null and b/assets/images/mixtral-of-experts-2.png differ
diff --git a/assets/images/mixtral-of-experts-3.png b/assets/images/mixtral-of-experts-3.png
new file mode 100644
index 0000000..9e29fe0
Binary files /dev/null and b/assets/images/mixtral-of-experts-3.png differ
diff --git a/assets/images/mixtral-of-experts-4.png b/assets/images/mixtral-of-experts-4.png
new file mode 100644
index 0000000..6494e2d
Binary files /dev/null and b/assets/images/mixtral-of-experts-4.png differ
diff --git a/assets/images/mixtral-of-experts-5.png b/assets/images/mixtral-of-experts-5.png
new file mode 100644
index 0000000..d4fa78a
Binary files /dev/null and b/assets/images/mixtral-of-experts-5.png differ
diff --git a/assets/images/mixtral-of-experts-6.png b/assets/images/mixtral-of-experts-6.png
new file mode 100644
index 0000000..b1b1ac9
Binary files /dev/null and b/assets/images/mixtral-of-experts-6.png differ
diff --git a/assets/images/mixtral-of-experts-7.png b/assets/images/mixtral-of-experts-7.png
new file mode 100644
index 0000000..f95270c
Binary files /dev/null and b/assets/images/mixtral-of-experts-7.png differ
diff --git a/assets/images/pixtral-12b-1.png b/assets/images/pixtral-12b-1.png
new file mode 100644
index 0000000..ae8b790
Binary files /dev/null and b/assets/images/pixtral-12b-1.png differ
diff --git a/assets/images/pixtral-12b-10.png b/assets/images/pixtral-12b-10.png
new file mode 100644
index 0000000..6bf54ea
Binary files /dev/null and b/assets/images/pixtral-12b-10.png differ
diff --git a/assets/images/pixtral-12b-11.png b/assets/images/pixtral-12b-11.png
new file mode 100644
index 0000000..a762e5d
Binary files /dev/null and b/assets/images/pixtral-12b-11.png differ
diff --git a/assets/images/pixtral-12b-12.png b/assets/images/pixtral-12b-12.png
new file mode 100644
index 0000000..56c2463
Binary files /dev/null and b/assets/images/pixtral-12b-12.png differ
diff --git a/assets/images/pixtral-12b-13.png b/assets/images/pixtral-12b-13.png
new file mode 100644
index 0000000..578bfce
Binary files /dev/null and b/assets/images/pixtral-12b-13.png differ
diff --git a/assets/images/pixtral-12b-2.png b/assets/images/pixtral-12b-2.png
new file mode 100644
index 0000000..7d0fc93
Binary files /dev/null and b/assets/images/pixtral-12b-2.png differ
diff --git a/assets/images/pixtral-12b-3.png b/assets/images/pixtral-12b-3.png
new file mode 100644
index 0000000..c3a4cc2
Binary files /dev/null and b/assets/images/pixtral-12b-3.png differ
diff --git a/assets/images/pixtral-12b-4.png b/assets/images/pixtral-12b-4.png
new file mode 100644
index 0000000..fbccf9d
Binary files /dev/null and b/assets/images/pixtral-12b-4.png differ
diff --git a/assets/images/pixtral-12b-5.png b/assets/images/pixtral-12b-5.png
new file mode 100644
index 0000000..b604a39
Binary files /dev/null and b/assets/images/pixtral-12b-5.png differ
diff --git a/assets/images/pixtral-12b-6.png b/assets/images/pixtral-12b-6.png
new file mode 100644
index 0000000..16e7ada
Binary files /dev/null and b/assets/images/pixtral-12b-6.png differ
diff --git a/assets/images/pixtral-12b-7.png b/assets/images/pixtral-12b-7.png
new file mode 100644
index 0000000..98efe1e
Binary files /dev/null and b/assets/images/pixtral-12b-7.png differ
diff --git a/assets/images/pixtral-12b-8.png b/assets/images/pixtral-12b-8.png
new file mode 100644
index 0000000..b698242
Binary files /dev/null and b/assets/images/pixtral-12b-8.png differ
diff --git a/assets/images/pixtral-12b-9.png b/assets/images/pixtral-12b-9.png
new file mode 100644
index 0000000..908b7f2
Binary files /dev/null and b/assets/images/pixtral-12b-9.png differ
diff --git a/assets/images/qwen3-technical-report-1.png b/assets/images/qwen3-technical-report-1.png
new file mode 100644
index 0000000..ba782c2
Binary files /dev/null and b/assets/images/qwen3-technical-report-1.png differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-1.png b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-1.png
new file mode 100644
index 0000000..e7f148a
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-1.png differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-2.jpg b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-2.jpg
new file mode 100644
index 0000000..0fad5ae
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-2.jpg differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-3.png b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-3.png
new file mode 100644
index 0000000..2c22b62
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-3.png differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-4.jpg b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-4.jpg
new file mode 100644
index 0000000..109bc24
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-4.jpg differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-5.jpg b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-5.jpg
new file mode 100644
index 0000000..e1cdd6e
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-5.jpg differ
diff --git a/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-6.png b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-6.png
new file mode 100644
index 0000000..f00d908
Binary files /dev/null and b/assets/images/raptor-recursive-abstractive-processing-for-tree-organized-retrieval-6.png differ
diff --git a/docs/architecture.md b/docs/architecture.md
index 533948f..6df5656 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -302,7 +302,9 @@ checks.
Five layers, each catching what the others cannot:
-1. **Unit tests** (`test/`): plugin functions and the lint, without a site.
+1. **Unit tests** (`test/`): plugin functions and the lint, without a site;
+ `test/js/` tests the search script's matching and ranking in node
+ (`node --test 'test/js/*.test.js'`, no packages).
2. **Source lint** (`script/lint-posts.rb`): front matter against
`_data/post_conventions.yml`, plus Markdown defects the built HTML no longer
shows (broken tables, unsafe math, text read as Liquid or HTML, bad fences).
diff --git a/docs/writing.md b/docs/writing.md
index 51d8979..388a95b 100644
--- a/docs/writing.md
+++ b/docs/writing.md
@@ -55,7 +55,7 @@ subtitle: "