I build AI systems that understand people before they automate the work.
I have cloned a doctor's brain into an AI model, built voice systems that replaced ElevenLabs with open-source tooling before that became the obvious move in 2025, and shipped AI products across chat, voice, documents, and workflow automation.
The pattern is always the same: understand how a person thinks, where the work slows down, then build the system around that. Not around a demo. Around the real job.
Most of my work sits at the edge of AI agents, RAG, voice, and document intelligence. I take messy ideas, half-working prototypes, and manual workflows, then turn them into tools people can use in the real world.
- Cloned a doctor's decision-making style into an AI model for healthcare-style reasoning and support workflows
- Replaced ElevenLabs-style voice infrastructure with open-source speech tooling before open-source voice became mainstream in 2025
- Built and shipped 20+ AI implementations across chat, voice, document processing, and automation
- Worked on systems handling 3,000+ page documents
- Built web-connected AI assistants with RAG and production integrations
- Turned slow manual document workflows into repeatable pipelines
- Contributed across 100+ repositories while building in public and for clients
- AI agents and chatbots that connect to real business data
- RAG systems for search, support, sales, and internal knowledge
- Document intelligence pipelines for long, messy files
- Voice tools, Telegram bots, Discord bots, and workflow automation
- Production backends with Python, FastAPI, Node.js, Docker, and cloud infrastructure
An AI system built to mirror a doctor's reasoning style, decision patterns, and knowledge workflow so users could get faster, more context-aware support.
A Kokoro-powered voice system built as an alternative to closed voice APIs. The goal was simple: own the voice layer, reduce dependency, and move faster.
A web-connected conversational AI platform built around RAG, sales workflows, and practical business use cases.
A document summarisation and intelligence pipeline for processing large files without making users dig through thousands of pages manually.
Infrastructure for fine-tuning and testing language model workflows at scale.
- AI: LangChain, LlamaIndex, OpenAI, Anthropic, Hugging Face, Groq
- Voice: Kokoro, TTS pipelines, open-source speech models
- Languages: Python, JavaScript, TypeScript
- Backend: FastAPI, Flask, Node.js
- Frontend: React, Next.js
- Infra: Docker, AWS, Azure, GitHub Actions
- Vector DBs: Pinecone, Weaviate, Milvus, ChromaDB
I like simple systems that do the job. Ship the first useful version, measure what breaks, then improve it until it feels obvious.
I care about the human part of AI as much as the technical part. If the tool does not help someone think faster, sell better, learn quicker, or save time, the model choice does not matter.
Building AI products that understand the work before they automate it.