> hello, world.

Osman Zain

AI & automation engineer

I'm an AI engineer who builds voice and chat automation for small businesses.

I design the whole thing, from the prompt to the phone call, and test it with evals and guardrails so it holds up on real customers, not just in demos. Recent work: GarageBot, a voice receptionist for auto repair shops that you can talk to live.

Open to AI / automation engineering roles

#Projects

What I've built, live for you to try.

GarageBot

Talk to it live ↗
problem
Independent auto repair shops lose bookings every week because nobody can pick up when they're under a car or the shop is closed. Industry data puts missed calls at 20–30% of inbound volume, and the average repair job at independent shops in Australia runs AU$250–400. If a shop misses just 3 to 5 bookings a week, that's roughly AU$39,000 to AU$104,000 a year. That range is my own arithmetic from those two figures, not a measured result for any one shop.
what I built
an AI receptionist for Australian auto repair shops. It answers phone calls around the clock, trained on each shop's own hours, services and prices, then hands the shop a job card with the caller's details. Fewer missed calls, fewer lost bookings, no extra staff.
  • Multi-tenant RAG on FastAPI: a separate ChromaDB collection per shop so one shop's data never enters another's answers.
  • OpenAI-compatible streaming endpoint used as the voice platform's custom LLM (Telnyx, then Vapi). Tuned prompts to keep spoken replies to about 15 words.
  • Tuned the live demo from real test calls: no diagnostic questions, one question per turn, re-asks when speech recognition mishears, and hangs up by itself.
  • Debugged real call failures from raw requests: caller identity fields, retried webhooks, and a flaky tunnel behind slow replies.
  • Python
  • FastAPI
  • ChromaDB
  • LangChain
  • Claude API
  • Vapi
  • Telnyx Voice AI

▶ try it live

Talk to the AI receptionist for a demo shop, "Aussie Auto Repair Shop". Ask about prices or book a service.

#Skills

llm / ai

  • Anthropic Claude API
  • Prompt iteration against evals
  • LLM-as-judge design
  • Guardrails & output validation
  • Structured output / tool use
  • RAG
  • ChromaDB
  • LangChain
  • Sentence-Transformers

automation & backend

  • Python
  • n8n
  • FastAPI
  • Flask
  • SQLite / SQL
  • Webhooks & idempotency
  • IMAP / SMTP
  • JavaScript

voice & telephony

  • Vapi
  • Telnyx Call Control
  • Telnyx Voice AI
  • WebRTC

data & ml

  • Pandas
  • NumPy
  • Scikit-learn
  • XGBoost
  • Model evaluation

quality & tools

  • Eval harness design
  • Offline tests with fakes
  • Prompt-injection testing
  • Git / GitHub
  • Tailwind CSS
  • Fly.io

#About

I build AI systems that deal with real customers, like a voice receptionist that answers a shop's phone. I care most about the details that decide whether they hold up day to day, like what happens when the model is wrong, when a caller interrupts, or when the same webhook arrives twice.

Small businesses are the customers I enjoy building for, because the result is easy to see: a call that used to go to voicemail now gets answered, and the details land where the team will see them.

#Contact

Hiring for an AI or automation role, or want to talk about a project?