Hackathon reports

Eval benchmarks and other HTML outputs from the team. Everything on main, plus what any unmerged branch adds or changes. To add one, put reports/<name>/index.html on any branch; see reports/README.md.

main

• IntelliJev Jev context evaluationTaanviir · 2026-09-23

  Jev + BM25 file retrieval in IntelliJev on 70 held-out Koog tasks: recall@10 0.69 vs 0.53 for BM25. Includes a separate eight-task agent pilot.

• IntelliJev context engine, explainedTaanviir · 2026-09-23

  The non-technical tour: the problem, the idea (Jev decides, the LLM writes), what it looks like, and the proof in plain numbers.

• IntelliJev local Laya evaluationaikram42 · 2026-09-23

  A frozen 30-task retrieval benchmark, exact-response caching and server resource evidence. Keyword search currently wins on retrieval.

• IntelliJev jury Q&ATaanviir · 2026-09-23

  The 30-second pitch, the numbers to remember, honest answers to likely jury questions, and what to do if the demo breaks.

• IntelliJev context engine wikiTaanviir · 2026-09-23

  Technical reference: Jev, BM25, sketches, why re-rank, fusion, stage 3, how the eval avoids fooling itself, IntelliJ internals, code map, glossary.

• IntelliJev
• IntelliJev pitch