Independent AI engineer · Istanbul · Remote
AI that worksin production,not just in a demo.
I build AI systems for companies: voice agents that answer your phones, search that actually finds things in your documents, agents that take repetitive work off your team, and computer vision for the counter and the shelf. One engineer, from the first call to the running system.
- > answer the clinic phone line in Turkish[ok] voice agent · first word in ~400 ms
- > find clause 4.2 across 3,000 contracts[ok] hybrid RAG · answer with source
- > recognise a simit without a barcode[ok] vision · runs on a CPU, 2nd of 400
- > turn a prompt into a 30s product video[ok] pipeline · Veo / Kling with fallback
Services
What I can build for you
Each of these is something I have already built and run myself. You get a working system with numbers on it, not a slide deck.
- [01]
Voice AI agents
An agent that picks up the phone, answers questions, books appointments and hands over to a person when it should. Turkish and English.
Call centres, clinics, rentals, service businesses
- [02]
Document AI & RAG search
Ask your contracts, policies and product docs a question and get an answer with the source. Hybrid search, re-ranking, access control.
Law, finance, insurance, support teams
Proof6-stage hybrid retrieval with re-ranking - [03]
Agents & workflow automation
LLM agents wired into your existing systems — email, CRM, ERP, spreadsheets — that do the repetitive back-office steps and log every action.
Operations, back office, e-commerce
ProofLangGraph pipelines, MCP tool registries - [04]
Computer vision
Recognise products without barcodes, count stock on a shelf, check a counter from one camera frame. Works on a CPU, learns a new item from a few photos.
Retail, bakeries, cafés, POS vendors
- [05]
AI features in your product
Add AI to the app you already have: an assistant, smart search, summaries, generation. Behind your API, with cost and latency limits you set.
SaaS teams without an AI engineer
ProofFastAPI / .NET / Node backends - [06]
Generative media pipelines
Prompt to storyboard to images to video, automated. Model fallback, rate limits and cost tracking built in.
Agencies, e-commerce, content teams
ProofMulti-model video pipeline (Veo, Kling, Gemini)
Selected work
Products I built and shipped
SpeakTableLive voice role-play rooms for language practice. Real people and AI personas share a scene while a private coach whispers corrections to each learner.
- 1.14 s
- voice pipeline p50, end of speech → first reply
- ≈ $0.50
- model cost of a 15-minute session
BarcodieA point of sale that recognises unbarcoded bakery items — simit, poğaça, croissant — from a single camera frame. Second place out of 400 at the Moka United AI Hackathon.
- 2nd / 400
- Moka United AI Ideathon & Hackathon
- 92.5–100%
- first-guess accuracy after 3 photos per product
litA campus-only social discovery app for university students, verified by school email. I built and run the backend — matching, real-time chat, ranking and the infrastructure.
- 370+
- backend commits, the large majority of the service
- UT-xxxx
- structured error catalog the app switches on
How we work
From first call to a running system
- [01]30 min
Free call
You explain the problem. I tell you honestly whether AI is the right tool and what it would roughly cost.
- [02]1–2 weeks
Prototype
A small, paid proof on your real data with a number we agree on up front: accuracy, time saved or cost per call.
- [03]3–8 weeks
Production
The real system: integrated with your stack, monitored, with fallbacks for when the model is wrong.
- [04]Ongoing
Handover
Docs, a runbook and your team trained. Optional monthly support and model updates.
Right now
- SpeakTable submitted to the App Store
- Published Barcodie model benchmark
- Barcodie pivots to a recognition API for POS
- 2nd of 400 at Moka United AI Hackathon
I co-run 4ourstack, a two-person product studio. We build our own apps and take on a few client projects a year.
Contact
Tell me about
your problem.
A few lines are enough. I reply within one working day — with questions, a rough plan, or an honest “AI is not the answer here”.