Isaacus open-sources a legal conflict detector: encoder-only pipeline reviews contracts in under 20s
Neon0asis · reddit · 2026-11-16
Isaacus open-sourced (MIT) a legal document conflict detector that uses small discriminative encoder models instead of generative LLMs for clause-level pairwise review.
Why not LLMs: pairwise clause comparison is quadratic — two 133-clause contracts mean 17,689 API calls; dumping files into Claude/GPT yields weak recall and hallucinations.
Pipeline: (1) parse documents into hierarchical clause trees, (2) embed clauses, (3) zero-shot legal classifier drops non-provisions, (4) shortlist top-3 matches with cosine similarity ≥0.5, (5) reranker scores relevance + NLI classifier scores entailment, (6) same-topic high conflict = deviation, no match = gap.
Longest document pairs ran in under 20 seconds. An optional LLM step only summarizes/redrafts findings, so it can't hallucinate conflicts into existence. Demoed on Bob Iger's 2011 vs 2022 Disney employment agreements, finding the 2011 version more favorable (e.g., no IP assignment clause).
More from Apps
- Obsidian Starter Kit v6: One Vault for Your Whole Life, AI Does the Work — dSebastien · 2026-11-16
- VEXO: free open-source tool batch-posts ComfyUI images to Pixiv, DeviantArt, Civitai, X and more — Mittishura · 2026-11-16
- ChatGPT voice dictation ditches old 10-minute limit, transcribes long input flawlessly — nptacek · 2026-11-16
- Anthropic Launches Claude for Manufacturing With 17 Skills and 4 Connectors — emmanuelvivier · 2026-11-16
- Pokemon-Style K5 Math Game Monster Math Released as Open Source — AIandDesign · 2026-11-16
- The Browser Company demos prototype blending full-duplex voice models with traditional GUIs — charliedeets · 2026-11-16