Featured Projects.
A collection of AI, machine learning, data engineering, RAG, and automation projects built from idea to implementation.
Financial Filings RAG Copilot
A production-style Generative AI application that enables users to ask questions over financial filings using LLMs, vector databases, document chunking, retrieval pipelines, and FastAPI services.
Automated Road Damage Detection System
A YOLOv8-based computer vision system for detecting and classifying road damages such as cracks and potholes from road images.
Financial Filings RAG Copilot
An AI-powered financial research assistant that answers questions from SEC filings using retrieval-augmented generation.
Data Quality Monitoring Platform
An automated data quality monitoring system that detects nulls, duplicates, schema drift, anomalies, and validation issues.
OPAL-AI Organ & Blood Matching System
A healthcare-focused matching platform for donor-recipient compatibility using FastAPI, Supabase, and intelligent matching logic.
Customer Booking Prediction Model
A classification model predicting whether a customer will complete a holiday booking using behavioral and travel-related features.
Multi-Agent Content Operations Engine
An autonomous multi-agent research and content system built with LangGraph, orchestrating Researcher, Writer, and Critic/Editor agents in a cyclic workflow to turn a raw topic into a polished, fact-checked report. The Critic/Editor agent enforces a style guide, checks drafts for hallucinations, and triggers rewrites until quality constraints are met, with a human-in-the-loop approval checkpoint before the final publish step. Exposed as a FastAPI service.
Autonomous AML & Fraud Investigation Agent
An end-to-end AML/fraud investigation pipeline triggered by Kafka transaction alerts, using asynchronous ingestion and enrichment to combine internal ledger data, OSINT, and corporate-registry lookups. Transaction networks are modeled as a knowledge graph in Neo4j using GraphRAG techniques to identify cyclical transaction loops, high-velocity layering, and cross-border structuring patterns. Instructor/Pydantic structured outputs draft regulatory-ready Suspicious Activity Reports, verified by a deterministic layer that cross-checks generated claims against raw transaction logs before human review.
AI Chatbot (LLM-Based Assistant)
A multi-turn AI assistant with memory handling and contextual responses built using LLMs and FastAPI, improving context retention by approximately 25–35%.
AI Voice Calling Agent
A prototype outbound voice agent built using LangGraph workflows, VAD, Whisper STT, LLM orchestration, TTS, MongoDB prospect context, and FastAPI/Zoom call control. Conversation flows, cancellation, playback accounting, MongoDB retries, dry-runs, and production-context rehearsals were all tested, with GPU/audio latency tracked and outbound acceptance supervised as a release gate.