cuesheet .
Open-source Python library that records LLM API calls once and replays them in tests. Drop-in pytest plugin, scrubs API keys, ships with a live web UI. Works with any SDK on httpx: Anthropic, OpenAI, Gemini, Mistral.

Python Engineer · AI Automation / Greece
Python backends, AI automation, and the plumbing between them. I use Claude Code as a daily driver across the full build cycle: planning, implementation, review. From northern Greece, working with startups that need someone who owns the whole stack.
I'm Giorgos. I write Python for a living, use Claude Code daily, and have spent the last few years moving between backend code and the networking layer underneath it. These days I lead a small team at Datawise (out of Atlanta, though I'm in Florina) that builds Python environments for training and evaluating AI models. Before that: e-commerce backends, satellite internet troubleshooting, and a lot of late-night customer calls about DNS.
Right now: leading a team at Datawise (Atlanta, working from Greece) that builds Python tasks used to evaluate and benchmark AI models. Containers, eval harnesses, and a lot of "why did the model do that" debugging.
Before that: e-commerce backends at Brdy., satellite internet support at Bigblu, and an internal Flask tool at Nova that cut hardware-replacement time in half.
I use Claude Code as a daily driver across the full build cycle: planning, implementation, review. Not as a shortcut but as a force multiplier for getting the hard parts done faster. It is also one of the models I evaluate at work, so I know where it is strong and where it needs a second pair of eyes.
Stack I reach for daily:
From Postgres up to the React you click on. Next.js shops, Flask APIs, the boring middle layer that holds it all together.
Python and PHP services that survive bad networks, ugly inputs, and Monday-morning load spikes. Tested. Logged. Observable.
Claude Code power user. I build with AI across the full cycle and can bring that into your workflow: LLM pipelines, eval harnesses, agentic automation, code generation setups for the repetitive parts.
Inherited code on fire? Networking weirdness nobody can repro? I like the unglamorous work. Most things have a root cause if you keep digging.
I write the migration myself. Indexes where queries hit. No surprise N+1s six months later.
Not just JWTs and HTTPS. RBAC, secrets handling, threat modeling - the stuff that bites when you don't.
Profile first, then fix. Most slow code is slow for one reason you can find in an hour.
Stripe to Postgres to the email job that fires at 9am. The integrations that make a product feel like one thing.
Tcpdump if I have to. The bug usually lives in the boundary you didn't think to check.
Built for the next 10x, not the next 1000x. Premature scale costs more than late refactoring.
Claude Code is part of my daily build cycle. I also evaluate and benchmark AI models at Datawise, so I know what these tools actually do. That combination means I use them where they help and skip them where they don't.
You send what you have - a doc, a deck, a Slack thread, even just a rough idea. We get on a 30-min call.
Within a week you get a written scope: what gets built, what doesn't, a timeline, and a fixed price. You decide.
Weekly Loom updates, the staging URL from day one, real code in your repo from week one. No black boxes.
Production deploy + 30 days of free fixes if anything breaks. You own everything, end of story.
Backend systems, internal tools, integrations, e-commerce builds, anything that needs Python or PHP done right. I'm best on projects that are 4-12 weeks of work for one person. I turn down enterprise consulting and pure-design jobs.
It depends on the project. Pricing is per project or per contract, scoped to what you actually need. Tell me what you're building and I'll send a fixed quote upfront, paid in milestones, no hourly surprises.
Usually within 2 weeks of the brief being signed off. If something is on fire I can sometimes move faster - say so in your message.
Always. I'm based in northern Greece, EET (UTC+2). I overlap with most of EU and the Americas East Coast for live calls. Async-first for everything else.
Standard. I sign mutual NDAs before we discuss anything proprietary. Your code stays yours, your data stays yours, no portfolio entry without your written OK.
Yes - I do a lot of work as the extra hands inside small teams. I write in your style, follow your conventions, leave the codebase better than I found it.
Got a question that isn't here? Just ask.
Services I rely on and recommend. Sign up through these links to get started, and I may earn a small credit at no extra cost to you.
Tell me about the project. I'll come back within a day with whether I'm the right fit, a rough timeline, and a starting price. No sales pitch, no contracts to read before we even talk.