B.Tech CSE (AI/ML) graduate building end-to-end AI systems — from deep learning models to deployed, multi-service full-stack products.
- B.Tech CSE (AI/ML) graduate (2026), focused on shipping real deployed systems over pure theory
- Built and deployed 6 end-to-end AI/ML projects — from a CNN+BiLSTM+Attention fraud detector to a full-stack job tracker with a Chrome extension
- Comfortable across the full stack: model training → FastAPI backend → React/Streamlit frontend → cloud deployment
- Interested in Deep Learning, RAG/multi-agent systems, and production ML infrastructure
- Open to AI/ML Engineering and Full-Stack roles — actively interviewing
Sequence-based fraud detection that analyzes transaction behavior over time instead of scoring transactions in isolation. A CNN → BiLSTM → Attention pipeline flags account-draining patterns that look harmless individually but suspicious as a sequence.
Result: Precision 0.91 · Recall 0.77 · F1 0.83 · ROC-AUC 0.99 (highly imbalanced data) · real-time FastAPI inference + Streamlit monitoring dashboard
Python TensorFlow/Keras FastAPI Streamlit Deep Learning
Event-driven hotel workflow platform that routes guest requests through pre-stay, in-stay, and post-stay agents, with a RAG-powered hotel FAQ assistant running on a local LLM (Ollama + Phi-3). Handles the full ticket lifecycle with auto follow-up and stale-ticket escalation, VIP prioritization from real guest profiles, policy-compliant review collection (no review-gating), and proactive owner analytics that surface recurring issues before they escalate — plus a daily manager digest email.
Python FastAPI Streamlit Ollama + Phi-3 ChromaDB (RAG) Qdrant SQLite
📂 Repo
A full-stack tracker for job-hunting freshers juggling 50+ applications across platforms. One-click capture via a Chrome extension (auto-fills from LinkedIn, Naukri, Internshala, Indeed), a React pipeline dashboard, Google OAuth, and PWA support for mobile.
FastAPI PostgreSQL Supabase React JWT Auth Chrome Extension PWA
A five-layer, rule-based candidate ranking pipeline that filters 100,000 profiles down to a top-100 shortlist with per-candidate reasoning — no LLM, no black box, fully explainable, under 5 minutes on CPU. Built for the Redrob India Runs Data & AI Challenge.
Python Pandas NumPy Streamlit
An awareness-first carbon footprint app — one short form instead of daily logging, real sourced emission factors (not made-up multipliers), and results explained through relatable comparisons rather than a guilt-driven number. Optional Claude-personalized reduction tips with a rule-based fallback. Scored 92.2/100 in Hack2Skill's PromptWars AI evaluation.
Python Streamlit Pydantic Anthropic API Pytest GitHub Actions CI
Predicts whether an IPO will deliver >10% listing gains using only pre-listing, publicly available data — strict no-leakage discipline, no Grey Market Premium. Trained on Indian mainboard IPOs (2010–2025); seven models benchmarked, with CatBoost coming out on top.
Result: ROC-AUC ~0.927 (best) · 0.90–0.93 across all 7 models
Python Scikit-learn XGBoost LightGBM CatBoost Jupyter
📂 Repo
Shipping deployed, end-to-end AI systems — not just notebooks.