I build backend systems, ML pipelines, and production-deployed applications. Currently in Semester 7 at MEC, focused on backend engineering, retrieval-augmented generation, and computer vision.

• 🏆 1st place F1 Score at PSCDL 2026 Computer Vision Challenge, conducted by Vehant Technologies
• 🔭 Building Studify — an RAG-based academic assistant with citation-grounded answers
• 🚀 Deployed production apps on DigitalOcean with Gunicorn + Nginx
• 🤖 Integrating LLMs (Groq, DeepSeek-R1, Ollama) into real systems

1st place F1 Score (0.40) · 3rd place Inference Time · Conducted by Vehant Technologies

Classical computer vision pipeline for detecting left-behind objects in fixed-camera surveillance video.

• Memory-efficient circular buffer architecture — reduced footprint 95% (3.4 GB → 186 MB)
• Full Python package with CLI, unit tests (pytest), CI/CD via GitHub Actions
• Live Gradio demo deployed on Hugging Face Spaces

Python OpenCV NumPy Pytest GitHub Actions Gradio Hugging Face Spaces

Full-stack · Team of 2 · 4 months · Deployed on DigitalOcean

AI platform for resume analysis, job matching, and real-time mock interviews.

• FastAPI backend deployed on Ubuntu VM (Gunicorn + Nginx)
• Real-time bidirectional communication via Socket.IO
• Cloud-hosted MongoDB for multi-user session management
• REST API tested with Postman; frontend in React + TypeScript

FastAPI Python React TypeScript MongoDB Socket.IO DigitalOcean Nginx Gunicorn

Backend Developer · Ongoing

Answers student questions strictly from their own course material — PDFs, slides, and handwritten notes — with full citation traceability.

• Hybrid retrieval: BM25 sparse + FAISS dense with Reciprocal Rank Fusion
• OCR pipeline: Tesseract (printed) + TrOCR (handwritten) + LLaVA-Phi3 (diagrams)
• Query expansion via HyDE using DeepSeek-R1 1.5B
• Token-level grounding scores — every answer is verifiable
• Secure: bcrypt, JWT, OAuth2, AES-256

FastAPI Python FAISS BM25 DeepSeek-R1 Ollama Docling TrOCR LLaVA-Phi3 Pytest

Backend Developer · Team of 4 · 3 months

Automobile insurance fraud detection combining ML and computer vision.

• YOLOv11-based visual damage assessment on accident images
• Ensemble ML model: SVM + Random Forest + XGBoost
• JWT-authenticated role-based access control (Admin / Surveyor)
• Groq AI API integration for automated PDF claim report generation
• Cloud-deployed FastAPI backend with PostgreSQL

FastAPI Python YOLOv11 PyTorch Scikit-Learn XGBoost NumPy Groq API JWT PostgreSQL

Languages Python Java SQL TypeScript JavaScript Dart

Backend & APIs FastAPI REST APIs Socket.IO JWT Gunicorn Nginx

Mobile Flutter Riverpod Dio Hive

ML & CV Scikit-Learn PyTorch YOLOv11 OpenCV NumPy FAISS Ollama

Databases MongoDB PostgreSQL

DevOps & Cloud GitHub Actions DigitalOcean Linux Docker Hugging Face Spaces

Tools Git GitHub Postman Pytest Gradio

• 🥇 1st position — PSCDL 2026, Vehant Technologies (1st F1 Score, 3rd Inference Time)
• 🤖 Participated in Win With AI 2.0 — IHRD Tharang 2026, NSDC MEC
• 🤖 Participated in Win With AI — Tharang 2024, NSDC MEC
• 📡 Content Team Member — IEEE MEC Student Branch 2025
• 🔬 NLP Training Programme — ICFOSS + Govt. Model Engineering College
• 🎯 Volunteer — MAGIC 5.0, IEEE MEC SB 2025