Projects

Work I'm proud to ship.

A peek at the AI systems I've built — what they do, the stack behind them, and the impact they shipped with.

  • Jan 2025 — Jul 2025

    01 / 06

    Voice-on-Call AI Platform

    AI-powered calling system with natural voice interaction. Sub-second real-time speech recognition and intent handling powered by Gemini.

    Reduced human operator dependency by 40% via automated voice workflows.

    GeminiFastAPIVITSTwilioWebSockets
  • Dec 2025 — Present

    02 / 06

    Sangamner AI

    Civic AI assistant for Sangamner — instant real-time chat over WebSockets with contextual replies based on citizen queries.

    <200ms response latency • 500+ daily public interactions.

    FastAPIGeminiWebSocketsRAG
  • 2024

    03 / 06

    Agentic Workflow Engine

    Scalable agentic workflow design for production automations — composable tools, planners, and retries with full observability.

    Cut backend latency by 30% across AI platform services.

    LangChainAgentsPythonDocker
  • Jul 2024 — Dec 2024

    04 / 06

    AI Face Surveillance & ALPR System

    Multi-tenant CCTV surveillance pipeline with sub-second face recognition and Automatic License Plate Recognition (ALPR). Features a multi-threaded camera stream manager, YOLOv8/v9 pipelines, InsightFace embeddings, Qdrant vector search, and Meta WhatsApp integration.

    Automated real-time attendance tracking and dispatched instant WhatsApp alerts for security breaches.

    FastAPIYOLOv8InsightFaceQdrantWebSocketsONNXPostgreSQL
  • 2024

    05 / 06

    AI-Powered WhatsApp Clone Backend

    A real-time WhatsApp clone backend engine featuring automated conversational AI agents. Supports WebSocket message relays, message database persistence, Redis-backed state caching, and Gemini API integration for automated customer support responders.

    Created an autonomous chat gateway with sub-100ms message relay latency.

    FastAPIWebSocketsGeminiPostgreSQLRedisDocker
  • 2024

    06 / 06

    AI Training Module with RAG System

    Enterprise knowledge training engine that enables Retrieval-Augmented Generation (RAG) over custom datasets. Supports automated document chunking, semantic vector index compilation in Qdrant, and custom context injection for tailoring model responses.

    Built visual board analytics to review document vector alignments and retrieval scores.

    PythonFastAPIRAGQdrantLangChainVector Embeddings