EDAI Document & Requirement AI — Progress Snapshot

---

Overview

Two complementary, production-ready prototypes are implemented:

• DOCUMENT-AI.v01: An insurance policy assistant leveraging FAISS-backed retrieval with LLM reasoning to answer policy questions, with Streamlit UI and FastAPI API.
• REQUIREMENT-AI.v01: An automated requirements authoring system using OCR + RAG + NLP to extract, validate, and prioritize requirements, with Streamlit UI and FastAPI API.

Both apps share a modular design, emphasize reproducible pipelines, and produce actionable outputs.

Highlights at a Glance

• RAG & Retrieval: FAISS vector search for relevant context across documents.
• OCR Support: Tesseract-based text extraction from scanned PDFs and images.
• NLP & Reasoning:
  • DOCUMENT-AI: LLM-backed policy reasoning and structured responses.
  • REQUIREMENT-AI: Parsing, classification (functional/non-functional), ambiguity/conflict detection, and MoSCoW prioritization.
• Outputs:
  • DOCUMENT-AI: Structured JSON-like decisions via API, interactive Q&A via UI.
  • REQUIREMENT-AI: DOCX requirements, Excel reports, and ready-to-import user stories.
• Interfaces: Streamlit UI for analysts, FastAPI backend for automation.

Project Status

• DOCUMENT-AI.v01

  • Core ingestion, chunking, embeddings, retrieval: ✅
  • LLM reasoning and sessionized flows: ✅
  • Streamlit UI and FastAPI API: ✅
  • Example scripts and docs: ✅

• REQUIREMENT-AI.v01

  • Ingestion (PDF/TXT/Image) + OCR: ✅
  • Embeddings + FAISS retrieval + candidate synthesis: ✅
  • NLP parsing, validation (ambiguity/conflict), MoSCoW prioritization: ✅
  • Outputs (DOCX/Excel/User Stories), Streamlit UI, FastAPI API: ✅
  • Offline-friendly defaults; LLM wiring via config possible: ✅

Quickstart

Document-AI (Policy Assistant)

• README: DOCUMENT-AI.v01/README.md
• Run UI:

  cd DOCUMENT-AI.v01
  # Backend (example from project README)
  cd app && uvicorn main:app --reload --host 0.0.0.0 --port 8000
  # Frontend
  cd ../ui && streamlit run app.py

Requirement-AI (Requirements Authoring)

• README: REQUIREMENT-AI.v01/README.md
• Run UI:

  cd REQUIREMENT-AI.v01
  streamlit run ui/ui_app.py

• Run API:

  cd REQUIREMENT-AI.v01
  uvicorn app.api:app --reload --host 0.0.0.0 --port 8000

Artifacts & Outputs

• Requirement-AI outputs (after a run) are saved under REQUIREMENT-AI.v01/out/:

  • requirements.docx: Formatted requirements document
  • requirements.xlsx: Detailed, filterable report
  • user_stories.txt: Plain text user stories for import

• Document-AI responses are served via API/UI with retrieved clause explanations.

Next Steps (Suggested)

• Wire configurable LLM providers for Requirement-AI synthesis (OpenAI/LangChain) with on/off toggles.
• Add domain-specific validation libraries (compliance/policy rule packs).
• Persist vector indexes and metadata stores for faster cold starts.
• Integrate JIRA export for user stories and requirements traceability.