How sajidghoto/misaq-ai works
Misaq AI is a research prototype for statute-grounded contract-clause analysis. Its principal workflow parses and processes a frozen statute corpus, chunks and indexes it, retrieves relevant provisions for clause analysis, then extracts and verifies citations against the corpus. Experiment and evaluation components support reproducible retrieval and generation studies; contract ingestion is a separate documented capability. The main workflow follows the README; some internal links and ingestion wiring are not shown in the excerpts, so those relationships are omitted or marked unverified.
Main entry points: generation/pipeline.py Clause generation · statute.py Statute corpus · parsing/pipeline.py PDF parsing · processing/pipeline.py Statute processing
Components 21
Corpus preparation
- Statute corpus
src/misaq/domain/statute.py - PDF parsing
src/misaq/parsing/pipeline.py - Statute processing
src/misaq/processing/pipeline.py - Section-aware chunks
src/misaq/chunking/chunker.py
Retrieval system
- Text embeddings
src/misaq/embedding/sentence_transformer.py - FAISS index
src/misaq/indexing/index.py - Dense retrieval
src/misaq/retrieval/retriever.py
Clause analysis
- Clause generation
src/misaq/generation/pipeline.py - Generation backends
src/misaq/generation/backends/base.py - Analysis prompts
src/misaq/generation/prompts.py - Replay cache
src/misaq/generation/cache.py - Citation extraction
src/misaq/verification/extract.py - Citation verification
src/misaq/verification/verify.py - Corpus registry
src/misaq/verification/registry.py
Contract ingestion
- Contract readers
src/misaq/ingestion/readers - Document structure
src/misaq/ingestion/structure/builder.py - Ingestion evaluation
src/misaq/evaluation/ingestion_eval.py
Research evaluation
- Retrieval metrics
src/misaq/retrieval/metrics.py - Experiment runner
src/misaq/experiments/runner.py - Domain contracts
src/misaq/domain
Other components
- Researcher
Connections 20
- Researcher → PDF parsing: starts parse
README.md - PDF parsing → Statute corpus: produces records
README.md - Statute corpus → Statute processing: processes corpus
README.md - Statute processing → Section-aware chunks: creates chunks
README.md - Section-aware chunks → Text embeddings: embeds text
README.md - Text embeddings → FAISS index: builds vectors
README.md - FAISS index → Dense retrieval: searches index
src/misaq/retrieval/retriever.py - Dense retrieval → Clause generation: supplies context
src/misaq/generation/pipeline.py - Clause generation → Generation backends: requests generation
src/misaq/generation/pipeline.py - Clause generation → Analysis prompts: assembles prompts
src/misaq/generation/pipeline.py - Clause generation → Replay cache: reads or writes
src/misaq/generation/pipeline.py - Clause generation → Citation extraction: extracts citations
src/misaq/generation/pipeline.py - Clause generation → Citation verification: verifies citations
src/misaq/generation/pipeline.py - Citation verification → Corpus registry: checks corpus
src/misaq/verification/verify.py - Corpus registry → Statute corpus: loads processed acts
src/misaq/verification/registry.py - Contract readers → Document structure: uses structure
- Researcher → Experiment runner: runs studies
README.md - Experiment runner → Retrieval metrics: computes retrieval metrics
src/misaq/experiments/runner.py - Researcher → Contract readers: ingests contracts
- Researcher → Ingestion evaluation: evaluates annotations