// Hello, I'm
I build production-grade AI systems — from medical speech recognition and RAG-powered clinical tools to autonomous multi-agent workflows and LLM-powered backends that run daily in healthcare and enterprise environments.
What I Do
End-to-end AI engineering — from medical AI deployment to autonomous agentic systems.
My Background
Experience, education, and the technical depth I bring to every project.
Portfolio
Production AI systems I've designed, engineered, and deployed.
Built a production e-prescription Flutter app where doctors speak and AI transcribes in real time. Deployed MedASR (medical speech recognition) and MedGemma (Google's medical LLM) locally, exposed both as REST API endpoints via FastAPI, and connected to the Flutter mobile app. Outputs two results: a transcribed prescription text and AI-generated clinical suggestions — with the doctor retaining full control to accept or reject.
End-to-end patient analysis platform for IMC Hospital. n8n pulls data from Microsoft SQL Server across 6 parallel streams (labs, radiology, pharmacy, discharge notes, vitals, pathology), custom JavaScript merges and structures it by admission, generates comprehensive HTML clinical reports, chunks and embeds them into Supabase/pgvector, and a Claude Sonnet AI agent answers physician queries via RAG in real time.
Architected a 4-agent CrewAI pipeline in Python — Researcher, Writer, Editor, and Manager — that autonomously researches topics via DuckDuckGo, drafts articles based on a style guide, polishes content, and emails the finished output. Integrated Groq API for fast LLM inference and implemented a production workaround for a CrewAI/Groq API compatibility bug.
AI-driven banking assistant that analyzes customer profiles and recommends the most suitable bank via email. Implemented a custom RAG pipeline with Supabase/PostgreSQL (pgvector) for semantic profile matching. Engineered an autonomous multi-agent state-machine managing full account-opening lifecycles via webhooks and WhatsApp Business API.
Multi-channel order ingestion gateway consuming Facebook, Instagram Graph, and WhatsApp Business APIs. Built a scalable Node.js/Express backend on Railway with JWT auth and strict input validation. Implemented LLM text classification to detect purchase intent from unstructured social feeds — reducing manual processing overhead by 100%.
Large-scale automated reporting workflow comprising 60+ interconnected nodes processing 500+ records daily across 60+ parallel data streams. Designed a highly concurrent distributed fan-out architecture with conditional routing and error-handling. Reduced manual reporting effort by 70% through scheduled triggers and automated delivery.
Automated disease surveillance system with 3 parallel alert workflows (Main, PDSRU, DDSRU). Scheduled triggers fetch live disease data via HTTP APIs, custom JavaScript nodes classify severity, then automatically route HIGH PRIORITY and STANDARD alerts to health authorities via Gmail and WhatsApp simultaneously — zero manual intervention.
Get In Touch
Open to remote and on-site opportunities. Let's build something exceptional together.
Whether you need an AI system built from scratch or want to automate complex workflows — I'm one message away.