Built a local-first observability and debugging platform for AI agents, making every LLM decision, tool call, retry, and failure visible through a real-time trace viewer — designed as Chrome DevTools + Datadog for LangGraph agents
Engineered a microservices architecture with Next.js frontend, Spring Boot REST backend, and FastAPI/LangGraph Python runtime, connected via WebSockets for live step-by-step trace streaming into PostgreSQL
Implemented multi-agent orchestration supporting 10 built-in agent types and 20+ LLM models across Ollama, OpenAI, Groq, Anthropic, and Gemini, with per-run model selection, side-by-side run comparison, replay, and cancellation
Developed an autonomous improvement pipeline including AI-powered accuracy evaluation, regression scoring, failure reason tagging (15+ reason codes), optimization advisor, and agent instruction patching; integrated Prometheus and Grafana for production monitoring with 8 live dashboard panels
Built a full-stack investment portfolio aggregator that ingests financial data from three independent sources (Orders, MF Central, Account Aggregator) with intelligent deduplication of overlapping records across sources
Engineered a FastAPI (Python 3.11) backend with layered service architecture for data parsing, aggregation, and analytics computation, exposing Swagger-documented REST endpoints with Pydantic response models
Developed a React 18 (Vite) frontend with reusable components, Axios-based API layer, React Context for global state management, and a comprehensive investment dashboard with portfolio analytics and visualizations
Containerized both services with Docker and Docker Compose for single-command reproducible deployment, supporting both local development and production environments
Built a high-performance trading system in C++ with multi-threaded WebSocket server for real-time market data streaming and order execution
Integrated Deribit Testnet API with OAuth2 authentication, enabling seamless placement, cancellation, and modification of spot, futures, and options orders
Implemented performance monitoring tools to capture WebSocket propagation, order processing, and API latency metrics for system optimization
Designed a command-line interface supporting live order book visualization, position tracking, and instrument management for efficient trading operations
Engineered a full-stack Python automation platform using Selenium and BeautifulSoup4 to systematically scrape founder data from Y Combinator and send personalized LinkedIn connection requests.
Implemented secure OAuth 2.0 integration with the Google Sheets API to automatically structure and store scraped company and founder data for centralized intelligence.
Developed advanced state management to intelligently handle connection request responses (Pending, Already Connected, Email Required) and log outcomes to a JSON file for analytics.
Designed a robust, menu-driven CLI architecture with modular components for batch selection, scraping, and connection management, ensuring scalability and ease of use.
Engineered a scalable microservices architecture with React, Spring Boot, Scala, and PostgreSQL, ensuring efficient data pipelines and persistent storage for summaries
Integrated an AI-driven summarization engine using Python FastAPI and Google Gemini LLM, enabling automated extraction of key insights from unstructured website content
Deployed containerized services with Docker and Helm, implementing orchestration strategies aligned with enterprise-grade cloud-native infrastructure standards
Delivered user-facing features for real-time summarization, historical search, and data retrieval, supporting intelligent information management and decision-making workflows
Designed and implemented a full-stack web application for real-time streaming, searching, and filtering of Docker container logs using WebSocket and REST APIs
Engineered a scalable backend with Node.js, Express.js, and MongoDB Atlas, ensuring persistent log storage and efficient query performance
Developed a React.js frontend with dedicated components for live logs, search functionality, and timestamp-based filtering, enhancing user experience
Deployed the system on Render and Vercel, integrating containerized log generation with remote streaming for robust and seamless operation
Built an immersive 3D haunted environment using Three.js, enabling interactive navigation with realistic lighting, fog effects, and atmospheric rendering
Designed detailed models including textured walls, doors, graves, and animated ghost characters, enhancing the overall spooky environment and user engagement
Implemented dynamic lighting effects with point lights, spotlights, and animated ghost lights to create eerie and responsive visual experiences
Optimized performance through texture mapping, geometry reuse, and efficient rendering techniques, ensuring smooth frame rates across devices
Built and published sportscli as a pip-installable Python CLI tool that streams live sports data — scores, standings, fixtures, and player stats — directly into the terminal without leaving the development workflow
Integrated three free public APIs (Lichess for Chess, cricketdata.org for Cricket, football-data.org for Football) covering 8 leagues including the UCL and FIFA World Cup, with automatic first-run API key prompting and secure config storage in ~/.config/sportscli/config.json
Architected a modular plugin pattern where each sport is a fully self-contained module (client, display, app), enabling new sports to be added with no changes to any existing code
Implemented an interactive configuration wizard (sports config setup) with partial-key masking for display and per-sport key management; released on PyPI with a clean build and upload pipeline via python-build and twine
Cloud-Based Fire Detection and Air Quality System with AI (Under Prof Bhaktha)
Designed an IoT-enabled system integrating fire and air quality sensors (MQ2, MQ135) with microcontrollers, enabling accurate hazard detection within 5–10 ft range
Developed a cloud architecture using MQTT and MongoDB Atlas for secure, real-time data ingestion, scalable storage, and centralized monitoring across multiple nodes
Built a MERN stack dashboard with live logs, data visualization, and downloadable reports, ensuring < 1s latency updates and user-friendly access to historical data
Incorporated AI-based anomaly detection with Gemini models and WebSocket-driven alerts, reducing false positives and enhancing overall system reliability