How p-kundan/ai-requirements-assistant works
Project type and purpose
• This is a local-first, Python-based requirements engineering assistant with a Streamlit GUI (“app.py”) on top of a collection of processing modules in src/.
• It ingests requirement documents (.docx, .pdf, .xlsx, etc.), uses a local LLM (models/mistral-7b…gguf via llama-cpp-python) to classify & score requirements, builds a traceability matrix, and renders interactive graphs & exports.
File structure
• Top-level:
• app.py – Streamlit entrypoint (frontend)
• models/ – Local LLM model files
• src/ – Core processing modules
• data/ – Sample documents & test data
• tests/ – Unit tests for modules (e.g., test_extractor.py)
• Key modules in src/:
• file_loader.py – Handles uploads, format detection, parsing
• extractor.py – Extracts raw text from parsed docs
• nlp.py – Classification, ambiguity scoring logic (calls LLM)
• validator.py – Validates requirement formats & rules
• traceability.py – Builds Stakeholder→System→Test Case links, network graph
• Config & dependencies: requirements.txt, uses streamlit, llama-cpp-python, common Python libs.
README for additional insights
• Confirms offline, privacy-first design; explains features: upload, classify, explain, score, trace, export.
• Mentions roadmap (future PDF reports, MBSE integration) but no external services currently.