I build AI tools and models, and I take on paid engineering work. If you need help running an open-weight model on your own infrastructure, working through a training pipeline, or fixing an inference bottleneck, that's the kind of problem I work on.
Run open-weight language, speech, or vision models on your own infrastructure. I can help with serving stacks, deployment boundaries, integration and operational issues.
Linux · GPUs · inference servers · on-premise environments
Help with datasets, preprocessing, reproducible training runs, model adaptation and evaluation. Scope depends on your data, compute budget and model licence.
PyTorch · datasets · training pipelines · evaluation
Profile the actual bottleneck before changing code. Work can include batching, memory use, model loading, throughput and latency, with comparisons against a defined baseline.
Profiling · benchmarks · accuracy checks · deployment costs
Implementation help, integrations, maintenance and troubleshooting for developer tools and AI workflows. I also build and maintain projects in public.
Python · TypeScript · CLIs · agent workflows
We start with a short call or email thread, then I propose scope, deliverables, timeline and price in writing. Formats such as a fixed-scope project, a time-boxed review or an ongoing support retainer are possible; details are agreed per engagement.
I use Claude Code and other coding agents to move faster on implementation, tests and documentation. Every change is reviewed and tested by me before delivery, and I verify results against a defined baseline. Your code and data are not shared with a model provider without your agreement.
I prefer linking to the actual repository or model card over making broad claims. These are things you can open and inspect.
A portable agent skill for working with Google Colab runtimes from terminal-based coding agents.
Source code & docs ↗A modernized PyTorch inference path for NISQA, with attention to batching, reproducible evaluation and checkpoint compatibility.
Implementation & tests ↗A published Turkish text-to-speech model based on F5-TTS. The model card documents its current alpha status.
Model card ↗A Chrome extension for saving and reusing system instruction profiles in Google AI Studio.
Source & install instructions ↗An agent skill for porting PyTorch models to JAX with numerical parity checks, validation budgets and recorded evidence.
Source code & docs ↗A PyTorch-free JAX inference port of NISQA with bundled verified weights, golden parity tests and a published cross-framework benchmark.
Source code & docs ↗A public Turkish question-answering dataset. Its dataset card sets out the licence and version details.
Dataset card ↗More releases and datasets on Hugging Face. I share technical observations on X (@HCSolakoglu).
Tell me what you're running, where it breaks or slows down, and what a good result would look like. I can say whether it's within my scope before either of us spends time on a call.
Paid engagements are scoped and quoted individually; see Service Terms. This website does not accept orders or payments. Provider: Hasan Can Solakoğlu, sole proprietorship, tax ID 7730560333.