System Registry / Projects

Agents, retrieval, and platforms built under real constraints.

A technical record of model-facing tools, retrieval systems, secure backend services, and full-stack fintech platforms.

Active Build Coding Agent

Coding Agent Workbench

A repository-aware assistant designed to inspect context, plan scoped changes, call tools, edit code, run checks, and report what changed with a traceable execution loop.

Context

Reads repository structure and relevant files before proposing a change.

Execution

Uses MCP and LLM tool calling to inspect, patch, and validate code.

Observability

Captures tool activity and verification steps through OpenTelemetry.

PythonMCPLLM tool callingGitDockerOpenTelemetry

GenAI MCP Server

Enterprise Data Chatbot

Built an LLM chatbot integrating Open WebUI with enterprise data through a custom Python MCP server. The system exposed tools for dynamic table discovery, field lookup, choice-value resolution, and schema-aware query generation.

Problem

Users needed natural-language access to operational data without giving the model unsafe execution power.

What I Built

Replaced arbitrary script execution with a structured JSON query interpreter supporting filters, aggregates, and dot-walking.

Security

Implemented ServiceNow OAuth 2.0 authorization code flow so API access stayed tied to the authenticated user.

Python FastMCP Open WebUI OAuth2 Schema-aware tools

RAG Hybrid Search

Policy Retrieval System

Built a retrieval pipeline for policy and knowledge content using Elasticsearch BM25, vector embeddings, document parsing, chunking, and user-level access validation.

Retrieval

Combined lexical search and embeddings to make document lookup more precise for enterprise knowledge use cases.

Ingestion

Used a Docling-based document parsing flow to transform source files into searchable chunks.

Deployment

Containerized the MCP server, Elasticsearch, and ingestion workers with Docker Compose.

RAG Elasticsearch Vector embeddings Docker Compose Access control

FinTech Full-Stack

TPRM Platform

Built a vendor lifecycle management platform covering pre-onboarding, vendor onboarding, risk assessment, due diligence, annual performance review, and regulatory data protection requirements.

Architecture

Spring Boot, Angular, MongoDB, Temporal workflows, authorization, and Kafka notification pipelines.

Integrations

Connected third-party screening services through OAuth2 client credentials via the enterprise auth gateway.

Compliance

Led PII encryption work including Aadhaar vault handling and PAN masking for sensitive data protection.

Spring Boot Angular MongoDB Temporal Kafka OAuth2

Academic ML

Research and applied machine learning foundations.

Website Phishing Detection

ML classification model for detecting phishing vs. legitimate websites using URL and page features.

View paper

Person Detection & Tracking

Computer vision pipeline with OpenCV for detecting and tracking people in video feeds.

Traffic Sign Detection

CNN-based image classifier for autonomous driving and traffic monitoring scenarios.