Senior AI Developer | Full Stack
AI systems from the databaseto the interface.
I build LLM systems, AI agents and voice AI together with their backend, data model and interface, and I run them in production myself. In software development since 2016, most recently inside the development teams of two US clients.
Available now. Permanent or freelance, 100% remote.
- 2016
- Developing software since my apprenticeship began
- 11,000+
- automated tests in pytest and Jest
- 5
- own products in live operation
- 2
- US teams I worked in from 2024 to 2026
One request, six layers, one developer.
A request travels from the interface to the database and back. I work in every one of these layers, and I know the seams between them, where most failures begin.
Interface
Web frontends with Server Components, 3D elements and a focus on Core Web Vitals and technical SEO. Mobile apps with local storage and localization.
- Next.js 16
- React 19
- TypeScript
- Tailwind CSS 4
- Three.js
- React Three Fiber
- React Native
- Expo
Server and APIs
Stateless services with streaming, webhooks and authentication. Integrations with partners and payment providers, hardened against limits, outages and duplicate calls.
- Python
- FastAPI
- Pydantic 2
- asyncio
- Node.js
- REST
- WebSockets
- SSE
- OAuth 2.0
- JWT
- Stripe
Data
Data models with schema design, indexes, migrations and row-level access rules. Vector search for retrieval, local databases for apps.
- PostgreSQL
- Supabase
- Row Level Security
- pgvector
- psycopg 3
- SQLite
AI
LLM pipelines and AI agents with tool calling, structured outputs, cost caps and evals against a gold set. RAG over pgvector, integration through MCP and real-time voice AI.
- Anthropic Claude
- OpenAI GPT
- Google Gemini
- Qwen
- AI agents
- MCP
- RAG
- Evals
- Pipecat
- Twilio
Quality
Tests are part of development, not an afterthought. Every production failure becomes a test case, and coding agents only work against acceptance tests.
- pytest
- Vitest
- Jest
- Playwright
- ruff
- ESLint
- Code review
Operations
Delivery and operation on my own Linux servers, with automatic TLS, CDN and deployments through the pipeline.
- Docker
- Linux
- Caddy
- Cloudflare
- Hetzner
- GitHub Actions
Experience
Three years of apprenticeship at a software company, five years as the sole technical owner of two businesses of my own, and since 2024 LLM systems in production, for two US clients and in my own products.
Own products
2025 to present
Senior AI Developer | Full Stack
Self-employed, own products and client projects, Oberschleißheim, remote
- LLM features and AI agents in Python and TypeScript on Claude, GPT, Gemini and Qwen, with tool calling, structured outputs, embeddings in pgvector and evaluation against gold data.
- Real-time voice AI for telephony with Twilio and Pipecat, with voice activity detection, speech to text, text to speech and response latency measurement.
- End-to-end ownership of several own SaaS products and platforms, from the data model through backend and interface to deployment and operations.
- AI-native development with several coding agents working in parallel, steered through specifications, acceptance tests and review, backed by over 11,000 automated tests.
- Websites for business clients in legal services, healthcare services and management consulting, multilingual and prepared for search engines.
Next.js 16, React 19, TypeScript, Python, FastAPI, PostgreSQL, Supabase, pgvector, Claude, GPT, Gemini, Qwen, Pipecat, Twilio, React Native, Docker, GitHub Actions
04/2025 to 08/2026
Senior AI Developer | Backend and LLM Systems (freelance)
Projectivity Inc., Silver Spring, MD, USA, remote in the development team of an IT consultancy
- Multi-stage LLM pipelines in Python with Anthropic Claude and OpenAI GPT that ran in production in client projects.
- Machine-reliable model outputs through structured outputs following JSON Schema, Pydantic validation before any further processing, and null values instead of invented data.
- Protected operations against provider limits and cost overruns with bounded concurrency, backoff with jitter on 429 and 5xx only, cost caps per call and per day, and model routing to smaller models where they sufficed.
- Checked every prompt and model change against a gold set with evals before adoption, with deterministic checks ahead of any model-based judgment.
- Delivered as FastAPI services in Docker with PostgreSQL, shipped through CI/CD with GitHub Actions.
Python, asyncio, FastAPI, httpx, Pydantic, JSON Schema, pytest, PostgreSQL, Docker, GitHub Actions, Claude, GPT
09/2024 to 02/2025
Senior AI Developer | LLM Integration (freelance)
Develop Zen, Richmond, VA, USA, remote in the development team of a SaaS provider
- Integrated LLMs into existing Python applications, with structured outputs as fixed data structures instead of free text.
- Prompt and context engineering with system prompts, few-shot examples and versioned prompts for comparable runs.
- Automated web data collection with Playwright and BeautifulSoup and image checks with OpenCV and PIL.
Python, Pydantic, JSON Schema, Playwright, BeautifulSoup, OpenCV, PIL
01/2022 to 08/2024
Founder & Software Engineer | Automation and AI
Own online retail, automated, Munich, remote, alongside a physical retail store
- Designed, built and ran the entire retail system alone, as an extensive Python codebase of several background services for listing, orders, shipping, customer communication and monitoring, running continuously.
- Product listing as a multi-stage LLM pipeline that normalizes supplier data, generates texts and attributes through language models, assigns the category and publishes only after passing validation. Missing details went into a review list instead of being filled in by the model.
- Image selection with a vision model, duplicate products detected through perceptual image hashes.
- Order processing through to the tracking number, designed to be idempotent so no order is placed twice.
- Shared interface layer for several sales channels and suppliers, with its own authentication, token renewal and rate limit per provider.
- Language model reply drafts for customer inquiries, special cases for approval (human in the loop). Watchdogs, logging and counters per account and channel.
Python, asyncio, SQLite, LLM APIs, vision models, REST, OAuth 2.0, JSON Schema, Linux
11/2019 to 12/2021
Founder & Software Engineer | Backend and Integrations
Own online retail, Munich, remote
- Developed and maintained the technical foundation alone, with automated handling of orders and shipping through partner and supplier interfaces.
- Implemented automatic reconciliation of order and shipping data in continuous operation, with token renewal without manual intervention.
- Regularly reconciled supplier product, price and stock data so nothing sold out was ever sold.
- Kept orders, products and shipments in an SQL database and analyzed them with SQL, running on my own Linux server with scheduled jobs.
Python, REST, OAuth 2.0, SQL, Linux, cron
09/2016 to 08/2019
Apprenticeship as IT Specialist for Application Development (Fachinformatiker)
Crealogix, Puchheim, apprenticeship at a software company, final grade 1.8 (German scale, 1.0 is best)
- Java, JavaScript, Angular and SQL, database design, tests with JUnit and Jasmine, Git and Scrum. Also internal training sessions for apprentices.
Java, JavaScript, Angular, SQL
Method
The model describes, the code decides.
Generated content only refers to existing records. The program makes the selection.
What code can check, no model checks.
Mechanical rules run before any model-based evaluation.
No model change without measurement.
Every prompt and model change runs against a gold set before it is adopted.
State lives in the database.
Services stay stateless and can restart without losing anything.
Every production failure becomes a test case.
That way the same breakdown is never built twice.
About me
I am a senior AI developer from Oberschleißheim near Munich. After my apprenticeship at Crealogix I spent five years responsible for the entire technology of two businesses of my own, and learned the whole craft there, from the data model through integrations to operations.
Since 2024 I have built LLM systems for production, first inside the development teams of two US clients, today in my own products, all the way to real-time voice AI on the phone. I am looking for work in which I own an AI system across every layer, as an employee or as a freelancer.
Tell me what you want to build.
info@akayguen.devA short email is enough. I send my resume as a PDF on request.