Projects
⚙️ Flagship Projects
These systems explore how AI can be deployed in structured, accountable ways. Each project focuses on document intelligence, retrieval architecture and decision-support pipelines designed for high-stakes environments.
Document Intelligence & Risk Detection System
A document analysis tool that processes long-form texts and returns structured summaries, clause insights and potential risk indicators. The focus is on producing consistent, usable outputs rather than generic AI summaries.
Features:
- Upload support for PDF, DOCX and TXT files
- Structured summaries instead of free-form text
- Clause detection and missing-element checks
- Risk flags with confidence indicators
- Clean JSON outputs for integration into other workflows
Tech Stack: Python, LangChain, OpenAI API, Flask
Semantic Search & Case Explorer
A semantic search system built to explore large legal text collections more intelligently than keyword search. It retrieves relevant material using embeddings and returns grounded, structured summaries.
Features:
- Case citation and free-text querying
- Embedding-based similarity search
- Structured breakdown of cases (facts, ratio, obiter)
- Context-grounded responses to reduce hallucination
- Simple dashboard interface for browsing
Tech Stack: Python, Embeddings, Vector Database, RAG pipeline, Dashboard UI
AI Intake & Classification Assistant
A data-driven document triage system that extracts structured information and prioritises cases using calibrated scoring logic. Designed to simulate high-volume intake environments with measurable decision outputs.
Features:
- Upload support for PDF, DOCX and TXT files
- Entity extraction and issue tagging
- Urgency scoring using rule-based logic with an ML baseline
- Priority bands (Low / Medium / High)
- Structured case storage for review and analysis
- Exportable reports (PDF/email)
Tech Stack: Python, FastAPI, scikit-learn, PostgreSQL
What These Projects Show
- Retrieval architecture using embeddings and vector databases
- Structured output design and schema enforcement
- Data-driven classification and calibrated scoring logic
- Basic ML modelling and evaluation (precision, recall, error analysis)
- Workflow-oriented system design with audit-friendly data storage
- End-to-end build from backend services to user interfaces