LegalisAI: Real Estate Legal Case Assistant

A retrieval-based legal assistant that surfaces similar past MahaRERA case outcomes, strong/weak points, and relevant RERA sections for a user's real estate dispute (Information Retrieval, not fine-tuned/trained on this data — see NOTICE.md).

---

🔧 Principal Architecture

1. Embedding model: intfloat/e5-base-v2 — pretrained sentence-embedding model, used as-is (no fine-tuning). See eval/ for retrieval-quality measurements.
2. Corpus: curated MahaRERA case summaries + RERA FAQ pairs (not published — see NOTICE.md).

🧱 Stack

• legalis_api/ — FastAPI backend: embedding model, precomputed retrieval index, retrieval logic.
• app.py — Streamlit UI, calls the FastAPI backend over HTTP. Run the API first, then this.
• eval/ — retrieval-quality eval harness (hand-labeled queries + Hit@k/MRR scoring). Run python eval/run_eval.py before and after any retrieval change to measure impact.

▶️ Running it

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt -r legalis_api/requirements.txt
pip install googletrans==4.0.0-rc1 --no-deps  # see note in requirements.txt

# terminal 1
cd legalis_api && uvicorn main:app --reload

# terminal 2
streamlit run app.py

Requires Data/ (gitignored) present at the repo root — see .gitignore. The embedding model downloads automatically from Hugging Face on first run.

---

🖥️ I/O Description

• Input: User-provided case description
• Output:
  • Relevant sections of law
  • Relevancy score of sections
  • Strong and weak points associated with the case
• Added Features
  • Language Compatibility (Eng/Hindi/Marathi)

---

⚠️ Disclaimer

This tool is informational only and does not constitute legal advice. See NOTICE.md for the full disclaimer, licensing (MIT), and model attribution.

📄 Note

The curated case dataset is not published in this repository — see NOTICE.md.