Henil C Alagiya

About

Hi, I am Henil. I am currently based in NJ, USA. I completed my MSCS from NJIT. I have experience working with various technologies including Full Stack Development, Salesforce CRM, and AI integrations with MCP and RAG. I enjoy building software, whether big or small, to solve real-world problems. My primary focus is on Full Stack AI Engineering.

Education

Work Experience

Projects

• Advanced RAG Chatbot System

  Python, LangChain, LangGraph, ChromaDB
  Problem Statement: In RAG systems, we mainly face a balancing act: too much context causes hallucinations, while too little fails to provide enough information. Additionally, standard context vectors often retrieve isolated snippets, missing the critical surrounding details needed for a complete answer.
  The Idea: To solve the context paradox, I used 'Context Padding'—retrieving small, precise vector matches but expanding the window to include surrounding text before passing it to the LLM. This gives the AI the 'full picture' without diluting the search accuracy.
  Solution:

  • Utilized LangChain to orchestrate the hybrid retrieval pipeline (dense + sparse search) and re-ranking logic for precise context fetching.
  • Implemented LangGraph to manage the conversational state, allowing the system to "remember" context and perform complex query rewriting.
  • Deployed ChromaDB as the high-performance vector store, optimized to handle the custom "context padding" embeddings and metadata filtering.
• LinkedIn Jobs Filter

  JavaScript, Chrome Extension API
  Problem Statement: Scrolling through LinkedIn felt like a waste of time—seeing the same 'Promoted' ads and jobs I'd already applied to over and over again. I just wanted to see fresh, relevant opportunities without the noise.
  The Idea: The browser knows what it's rendering. I decided to inject a script that 'watches' the page build process. As soon as a job card appears, the script checks its properties (like 'Promoted') and removes it from the code often before it's even painted to the screen.
  Solution:

  • Utilized MutationObserver API to monitor the DOM changes in real-time, removing ad elements before they are painted to screen.
  • Implemented Vanilla JavaScript filtering logic to inspect job properties against user-defined criteria (Remote/Hybrid/Easy Apply).
• Apple Notes MCP Server

  Python, AppleScript, MCP
  Problem Statement: I love using AI to organize my thoughts, but it was frustrating that my local LLMs couldn't actually see or edit my Apple Notes. I wanted a way for my AI agents to seamlessly manage my personal knowledge base right where I keep it—on my Mac.
  The Idea: Apple Notes acts like a walled garden to outside software, but it speaks 'AppleScript'. I built a translation layer: a Python server that accepts modern JSON commands from the AI and 'translates' them into the native AppleScript commands that the Notes app understands.
  Solution:

  • Utilized FastMCP framework to expose Apple Notes functionality as a standardized Model Context Protocol server for agents.
  • Implemented AppleScript bridges to execute native macOS commands, enabling Python to "drive" the local Notes application.
  • Integrated Pydantic for robust input validation, ensuring that agent commands match the strict schema required by the tools.
• Google Sheets MCP Server

  Python, Google Sheets API, MCP
  Problem Statement: Connecting AI agents to spreadsheets usually felt clunky or limited. I wanted my AI to do more than just read cells—I needed it to truly 'drive' Google Sheets, handling complex data tasks autonomously like a real analyst.
  The Idea: Spreadsheets are just data grids with rules. I treated Google Sheets not as a UI, but as a set of callable functions. By wrapping the Google API in an agent-friendly protocol (MCP), I gave the AI specific tools to 'see' and 'touch' the data directly.
  Solution:

  • Utilized FastMCP to wrap 25+ Google Sheets API endpoints into agent-ready tools for read, write, and formatting operations.
  • Implemented Service Account Authentication to ensure secure, non-interactive access suitable for production backend environments.
  • Integrated Pydantic models to strictly validate data types before sending requests to the Google API, minimizing runtime errors.
• QuantBot

  Python, Pandas, TA-Lib
  Problem Statement: I realized that no matter how fast I reacted, I couldn't beat the market's speed or ignore my own emotions during volatile trades. I built QuantBot to take 'me' out of the equation—executing strategies instantly and purely based on data, 24/7.
  The Idea: Speed comes from removing friction. I removed the human bottleneck by building a 'headless' listener. It connects directly to the data stream, and purely mathematical logic decides to buy or sell—skipping the visualization step entirely to save milliseconds.
  Solution:

  • Utilized Pandas & NumPy to vectorise market data processing, enabling the system to evaluate strategies across millions of data points in milliseconds.
  • Implemented Event-Driven Architecture to decouple data ingestion from execution logic, ensuring sub-second reaction times.
  • Integrated TA-Lib to compute real-time technical indicators (RSI, Bollinger Bands) dynamically as new ticks arrive.
• QuickTrade

  Django, Python, PostgreSQL, Fyers API
  Problem Statement: In scalping, every millisecond counts. I found myself missing perfect entry points just because I was too busy calculating strike prices and fumbling with order forms. I needed a tool that could keep up with my thinking speed.
  The Idea: In scalping, thinking time is losing time. The idea was to 'pre-load' the gun. The system constantly calculates the correct strike price in the background, so the 'Buy' button is always armed with the perfect order parameters, requiring zero setup from the user at the critical moment.
  Solution:

  • Utilized Fyers Data API to subscribe to live WebSocket feeds for real-time index tracking and ATM strike calculation.
  • Implemented a Unified API Adapter in Python to normalize order execution endpoints across different brokers (Zerodha, Upstox).
  • Engineered the One-Click Execution Engine to pre-construct API payloads, eliminating manual form-filling latency.

Certifications

• Introduction to Model Context Protocol
  Anthropic · Oct 2025
• Salesforce Certified AI Associate
  Salesforce · Oct 2024