Faizan J.

I'm a full stack AI developer. I build RAG systems and computer vision models, then build the product around them: backend, frontend, database, deployment. Twelve projects in my portfolio, including RAG, computer vision, document extraction and voice AI. 🔧 Work I've done 📚 Paid report product built on a RAG pipeline over a Chinese-language source book. Free preview, then full bilingual PDF reports sold through Stripe. Admin dashboard for content, exports and usage stats. Next.js, Supabase, Gemini API, live on Vercel. I led the AI development on this. 👁️ Diagnostic support tool that helps eye doctors read OCT scans. Classifies four conditions (CNV, DME, drusen, normal) and shows a Grad-CAM heatmap so the doctor can see which region influenced the prediction rather than relying on the classification alone. Auto-generates a patient PDF with a plain language explanation. PyTorch ResNet-101, React, Flask, Supabase, with separate doctor and patient portals. 📄 Bank statement tool that turns PDFs into numbers an analyst can work with. Detects which bank a statement came from and applies a parser written for that format, then reconciles against closing balance so the totals tie out. Transaction rules, custom filters, monthly aggregates, PDF export. Next.js 14, TypeScript, PDF.js, Recharts. 🚦 Traffic monitoring on CCTV feeds. Counts and tracks cars, buses, trucks, motorcycles and pedestrians, flags congestion, reports flow over time. YOLO, OpenCV, Python. 🎙️ Voice agent you can talk to in a browser tab, no install or plugin. 100+ character personas with distinct voices and behaviours. Real-time speech in, speech out over WebRTC and WebSockets. FastAPI, OpenAI GPT, Docker, deployed on Railway with CPU and GPU support. ⚙️ What I work on RAG and agents: retrieval over documents, databases and APIs, vector search, tool calling, domain-specific AI workflows. Computer vision: detection, tracking, classification, segmentation, explainability. Voice AI: real-time conversational agents, STT and TTS pipelines, WebRTC. Shipping it: FastAPI, Flask, Next.js, Supabase, PostgreSQL, Docker. Vercel and Railway. You get a running system and documentation of how it was deployed. 💬 Most of my work involves messy inputs where an off-the-shelf model performs badly without adjustment. Send me the problem you're dealing with, and I'll come back with an approach, the architecture I'd use, and a cost estimate. If I think what you're asking for isn't feasible, I'll tell you before you spend money on it.