// Python · LangChain & RAG · FastAPI
AI-focused Computer Science graduate and ECPC competitive programming contestant with hands-on experience building and deploying Generative AI and Retrieval-Augmented Generation (RAG) applications — multi-model fallback systems, async APIs, and production deployments that bridge machine learning with backend engineering.
⚡ Powered by LangChain + Gemini AI
Production-ready RAG pipeline ingesting CV, LinkedIn, and GitHub data into a FAISS vector store, with per-session conversation memory, IP-based rate limiting, and secure admin endpoints. Ships through a GitHub Actions CI/CD pipeline to Hugging Face Spaces (Docker + Git LFS) and is integrated live into this very portfolio — try the "Ask AI" widget above.
End-to-end job-matching pipeline that scrapes live listings (Wuzzuf) via Playwright, embeds resume and job data with ChromaDB + sentence-transformers, and scores fit through a 3-stage filter — embedding similarity, LLM scoring, then threshold. Runs a multi-provider LLM fallback (Groq Llama 3.3 70B → Gemini → local Qwen via Ollama) with automatic quota/cooldown handling, and returns an actionable verdict (Apply / Improve then Apply / Skip) with a gap analysis synced to Notion.
Freelance-built adaptive learning platform: personalized learning paths, AI-driven MCQ generation, and live coding assessment. Features a context-aware AI tutor with automated grading and multi-language code evaluation, plus retry logic and multi-model fallback for reliability. Deployed to production on Railway.
Secure mobile voting system using blockchain for tamper-proof records, with AI facial recognition (DeepFace) for voter identity verification and anti-spoofing, plus an OCR-based ID verification pipeline with database cross-checking. Face-matching and ID-matching services deployed to AWS (EC2) behind a unified API.
Deep learning model classifying 7 skin diseases using MobileNetV3Small pre-trained on ImageNet. Medical-grade augmentation and class-weighted loss improved Melanoma recall by ~18%, reaching 81.6% test accuracy and 0.984 AUC. Exported to TFLite (4 MB, <1ms inference) for mobile.
Interactive game where a warrior navigates an arena full of enemies, using A* pathfinding, grid-based obstacle detection, real-time combat, and a scoring system.
Completed a structured summer internship focused on data analysis, visualization, and core ML concepts.
I'm open to new opportunities, collaborations, and interesting projects — especially in Generative AI, RAG systems, and backend engineering. Feel free to reach out!