Building production AI systems for enterprise clients across India and Malaysia. Multi-agent platforms, NL-to-SQL over 200K+ publications, AI recruitment pipelines — 99.9% uptime, 1000+ daily automated workflows.
MERN stack healthcare platform. MongoDB query optimisation cut dashboard load from 3s → under 1s. API validation reduced invalid requests by 80%.
./experience -v
A log of professional history and built systems
Full Stack Developer · Applied AI
- ->Built and deployed several AI features for clients across India and Malaysia.
- ->Created a custom workflow platform with Next.js and FastAPI that acts like a smart router, sending over 1,000 tasks daily to different specialized AI agents.
- ->Added a natural language search to a 200,000+ publication database using LangChain. I used Redis heavily here to keep the query times down.
- ->Automated resume screening using spaCy for text parsing and GPT-4 for scoring candidates.
- ->Trained a scikit-learn email classifier to categorize incoming messages and draft automated replies.
- ->Shifted a lot of our heavy API data aggregation to PostgreSQL and added Redis caching, which noticeably improved response times.
- ->Set up deployments across AWS, Vercel, and Render using Docker.
Full Stack Developer Intern
- ->Worked as a full stack developer on a healthcare platform using the MERN stack.
- ->Wrote Zod schemas for all our APIs to catch bad requests early.
- ->Added indexes to MongoDB and profiled slow queries to drop the main dashboard load time from 3 seconds to under 1 second.
prince@macbook:~/skills$pkg list --installed --icons
System capabilities and installed technical packages
./projects -l
Active repository list and deployed environments
Mental Health AI
Fine-tuned a Qwen-3B model entirely on Apple Silicon just to see how far I could push local hardware. I scraped 65,000 Reddit posts, cleaned the dataset, and generated synthetic replies using Ollama. It serves as both a 14-class symptom classifier and an empathetic chatbot. Served via FastAPI.
Multi-Agent Workflow Platform
Our production platform for running different specialized AI agents. You type a prompt, and a routing engine decides which agent should handle it (drafting emails, managing calendars, processing docs). It runs hundreds of automated tasks daily.
NL-to-SQL Media Monitoring Chatbot
A chat interface connected to a 3.6 GB MySQL database of publications. Users ask questions in plain English, and LangChain writes the SQL to fetch the answers. I added Redis caching so frequent queries load instantly.
AI Recruitment Screening System
An internal tool for HR to speed up hiring. It uses spaCy to extract context from a resume rather than just matching keywords, and then GPT-4 scores the candidate against the job description. It processes hundreds of resumes in seconds.
Delhi PM2.5 AQI Forecasting
My final year university project. I wanted to understand the math behind backpropagation, so I built a neural network completely from scratch in pure NumPy (no PyTorch or TensorFlow) to predict Delhi's air quality using 6 years of sensor data.
Intelligent Proctoring Framework
A real-time browser-based exam monitoring tool. It tracks where the user is looking and checks for multiple faces or objects to detect cheating. Mostly an excuse for me to play around with face-api.js and MediaPipe.
./connect_agent.sh
Initialize secure connection protocol