BYOK AI Chat
Chat with multiple AI model providers (OpenAI, Anthropic, Google) using your own API keys. Sync across devices, prompt library, up to 10× cheaper than standard subscriptions.
chat.juszczyk.it →I build AI systems that don't need babysitting at night.
You have a working business. You want AI where it actually makes sense — no full rewrites, no promises you can't measure. Architecture audit, finding where a model or agent shortens a process or cuts cost. Then integration with guardrails, quality evaluation, and a production rollout.
We build a product from scratch with AI at its core. A design that holds under production traffic and won't torch your model bill. Multi-provider setup, bring-your-own-keys where it makes sense, caching, fallbacks, model quality evaluation.
Prototype worked, production is breaking. Hallucinations, runaway costs, unpredictable behavior. Diagnosis, recovery plan, and an honest answer to one question: salvage what's there, or start over.
AI in a prototype is magic. AI in production is engineering. What counts is the system that does its job every day — not just on stage.
— Rule no. 1Chat with multiple AI model providers (OpenAI, Anthropic, Google) using your own API keys. Sync across devices, prompt library, up to 10× cheaper than standard subscriptions.
chat.juszczyk.it →Photo and video compression entirely in the browser. Full privacy — files never leave your device. Built on WebAssembly and FFmpeg.
compresso.juszczyk.it →Quick notes with AI summaries. Capture a thought in a second, get a clean digest back. In progress.
soon →I embed AI in existing systems, design architecture for new AI-native products, run audits, and rescue projects where the prototype worked but production is breaking. Specialty: language models (LLMs) that don't hallucinate in production and don't burn through your budget in a weekend.
Six years of production engineering in banking (BNP Paribas, Credit Suisse). Working with AI in production since 2023, AI Champion at Synergy Codes since 2024. First exposure to machine learning during my studies (Wrocław University of Science and Technology, MSc Eng. in Computer Science, 2017–2022). AI Devs 4 certified.
I work with multiple model providers — OpenAI (GPT), Anthropic (Claude), Google (Gemini), and local models (Llama, Mistral). The choice depends on the project: answer quality, data privacy, cost and infrastructure constraints. I often design multi-provider setups to avoid lock-in.
A project audit usually takes 1–3 weeks. Embedding AI into an existing system runs 4–12 weeks, depending on scope. Architecture and build for a new AI-native product takes 8–24 weeks. Every project starts with a short conversation and diagnosis. The concrete timeline comes after that.
Yes. I work remotely with clients across Poland and selectively internationally. In-person meetings around Szczecinek if the client prefers.
Per-project pricing after the first conversation and scope review. Audits have a fixed rate. Implementation and architecture — priced by scope and risk, not hours.
First a 30-minute conversation: context, problem, goal. Then diagnosis (1–3 days), a concrete proposal with scope and price, and execution in iterations with visible progress.