Agentic AI, large language models, retrieval-augmented generation, and computer vision — turned into working systems that solve real problems end to end.
I'm an Integrated B.Tech + M.Tech graduate in Artificial Intelligence and Robotics from Gautam Buddha University. My work sits where applied machine learning meets system design — building assistants that can reason over context, hold long-term memory, and act through multiple coordinated agents.
My academic focus is community detection in complex networks, spanning classical algorithms through graph neural networks. The same instinct — finding structure in messy, connected data — runs through everything I build, from RAG pipelines to computer-vision systems.
Across the last 3+ years, I've turned that instinct into working software — leading a small development team, co-authoring published research, and shipping deployed AI projects end to end.
I'm currently open to AI/ML Engineer and Python Developer roles where I can ship intelligent, data-driven products.
An advanced AI platform built around multi-agent reasoning, long-term memory, and retrieval-augmented generation. It handles document, vision, and voice inputs, and automates workflows by coordinating specialised agents — deployed as a live application.
A research-analytics platform that detects collaboration communities by building co-authorship networks from live OpenAlex publication data. It applies the Louvain algorithm to surface research communities and rank influential researchers, presented through an interactive dashboard — deployed to production on AWS.
An end-to-end machine-learning web service that predicts residential property prices across Indian cities. Built on a cleaned dataset of 64,000+ listings (30 cities, 3,000+ localities) — with careful removal of rental, mislabeled, and duplicate data, and normalized schemas — it serves a HistGradientBoosting model (scikit-learn Pipeline, log-transformed target) through a FastAPI REST API and a custom interactive frontend. Containerized with Docker and shipped with a GitHub Actions CI pipeline.
A pharma field-rep screen to log doctor (HCP) visits two ways — a structured form or a conversational AI panel. A LangGraph agent (router → tool → responder) turns plain-English notes into structured CRM writes across five tools, both panels sharing one backend.
A customer-complaint module for a pharmaceutical Quality Management System. Upload a complaint (PDF / DOCX / TXT / EML) or paste text; a LangGraph agent extracts 11 fields, assesses risk severity and priority, auto-fills the form, and saves to PostgreSQL — with a built-in AI Q&A chat.
An end-to-end pipeline that turns invoice emails into approved database records: Gmail → PDF extraction → a FastAPI + Gemini service returning structured JSON → Airtable, with a human-in-the-loop approval step. Orchestrated as two n8n workflows.
A data pipeline for a safe AI Vedic astrologer — validates chat structure and safety rules, generates new safe Hinglish astrology conversations, and scores response quality with an LLM judge (safety, warmth, honesty, helpfulness), wrapped in a Streamlit live-chat dashboard.
A production-grade URL audit API returning health and SEO signals (status, response time, title, meta, heading/image hygiene, links, HTTPS, security headers). Built for production: SSRF protection, timeouts, a concurrency cap, per-client rate limiting, caching, structured logging, tests, and CI.
A sentiment-analysis project that classifies product reviews as positive, negative, or neutral at 88% accuracy, comparing Naive Bayes, SVM, and Logistic Regression.
A real-time AI attendance system using facial recognition, reaching 95%+ accuracy with an OpenCV pipeline and a Flask back end.
An exploratory analysis (Primetrade.ai internship assignment) of how trader performance on Hyperliquid relates to the Bitcoin Fear & Greed Index — joining ~211,000 trades with daily market sentiment to surface actionable patterns.
Co-authored with Dr. Maneet Singh. A survey of traditional, probabilistic, and deep learning-based (GNN) community detection techniques — covering overlapping communities, temporal networks, explainability, and hybrid frameworks.
Fine-tuning a small open-source language model with QLoRA on a focused dataset — adapting an LLM for a specialized task on free GPU.
Open to AI/ML Engineer and Python Developer roles, and to collaboration on applied AI work.