About Me

Three Gates Before You Spend the Budget

30-minute talk opening the Startup Community Poznań meetup on fast AI validation. Three gates, two days each, that kill or pass an AI idea before the budget goes out. Gate 1 — is AI even the right mechanism: deterministic rules mean a form, no data means collect data first, full-text search means full-text search; then the ladder prompt → prompt with context → RAG → workflow → agent, with 150,000 policy PDFs a month at Insly as the workflow case (Claude Haiku 4.5 on Bedrock EU, 98.5% extraction accuracy, $5,500 a month). Gate 2 — does it close on paper: users × queries × cost per query, so 4,000 queries a month runs ~$0.60 on an open model hosted in the EU, ~$70 on a frontier model and $942 on a 24/7 cloud GPU, a 250× spread for the same feature, plus the hidden costs nobody counts (two-page system prompts, retries, growing chat history, embeddings in a loop, 30–60 s cold starts). Gate 3 — does it work on MY data: general models answer confidently and wrongly on a closed domain, so the only credible test is your own — 30 questions, expected answers, scoring, a hallucination counter, pass threshold set before the test. Our result: a 7B model beat an 11B one (0.949 vs 0.915, ~1.5× faster end to end) and the harness caught a build injecting junk tokens that manual clicking missed. Punchline: three RAG prototypes, two weeks each, two kill decisions, production on the third — six weeks instead of a 6+ month R&D cycle.

AI Models, Hermes & PCSS + From Prompt to Agent

Two 15-minute talks at the two-day Future_Lab (PARP Future Camp, Ryn Castle). First: what an LLM actually is without the transformer architecture, what it is genuinely good at, what it never guarantees (truth, currency, full context, responsibility), and Hermes plus PCSS models as the shared working environment. Second: one task walked through three levels — a good prompt (goal, context, constraints, expected result), then a skill, then an agent — arguing you do not start by building the agent. 40 participants, running problem to prototype to working PoC to demo.

AI in Academic Research for PhD Students

90-min practical workshop for non-technical PhD students (law, humanities, social sciences) at UAM. Local Bielik 4.5B via LM Studio: OCR cleanup, audio transcription, GDPR anonymization, RAG over personal PDF library. All data stays on-device.

Eskadra Bielika - Mission 2 — Sovereign RAG with Bielik & Google Cloud (2nd edition)

Hands-on workshop on building production-ready RAG systems with Bielik (Polish open-source LLM) on Google Cloud Run. New edition with updated presentation covering hallucination detection, guardrails, chunking strategies, HYDE, RAPTOR, and RAG evaluation.

PoznAI Bielik — Hands-on Workshop

A 7-hour workshop (10:00-17:00) putting Bielik, the Polish open-source LLM, in the hands of non-technical students through the PCSS API, with no local installs. It opened with three live ChatGPT-versus-Bielik demos: a rizz check, a terms-and-conditions translator and "Janusz mode". 28 participants in 7 teams then picked a use case and one of four starters (playground, Python pipeline, multi-step, Node UI) and built their own thing with mentors rotating between tables. Every team demoed. Winners, voted by the room: smart flashcards (Qwen reads a photo of your notes, Bielik turns it into questions), a cash-flow analyser for listed companies, and an "acid barometer" scoring relationship conversations. Organised by Kanał Studencki and PCSS in Poznań.

Eskadra Bielika - Mission 2 - Sovereign RAG with Bielik & Google Cloud

Hands-on workshop building a sovereign RAG system on Polish Bielik 4.5B LLM and Google Cloud (Cloud Run + BigQuery Vector Search). Covers model selection, embeddings, chunking, hardware trade-offs (L4 GPU, quantization), cost models, and lessons from production OWU AI pipeline at Insly.

Vibecoding - Rise of Agents. End of developer era?

Architecture patterns validated in production for AI-assisted development. Five proven patterns: Bounded Context Enforcement, Contract-First Development, Security Checklist Prompts, Performance Prompt Patterns, and Saga Pattern. Real case studies from insurance domain.

Effective Scaling of Monolithic Applications in PHP

Practical strategies for scaling PHP monolithic applications. Benchmarking, optimization techniques, and migration patterns to handle growing traffic without full microservices rewrite.

Inbox/Outbox Pattern - Reliable Inter-System Communication

Reliable inter-system communication patterns. Implementing Inbox/Outbox for guaranteed message delivery, handling distributed transactions, and ensuring data consistency in microservices architecture.

High performance PHP8 at Scale

Benchmarking PHP8 performance improvements. Siege load testing methodology, Docker containerization strategies, and real-world performance optimization case studies with Symfony Demo App.

Miserable Microservices Misconceptions

Common pitfalls and misconceptions when adopting microservices architecture. When monolith is the right choice, distributed monolith anti-patterns, and realistic assessment of microservices trade-offs.

Idempotency - The Key to Reliable Systems

Understanding idempotency in distributed systems. CQRS patterns, message queue handling with Kafka and RabbitMQ, and building resilient APIs that gracefully handle retries.

Technical Debt Reduction - How to Stay Sane

Practical refactoring strategies for legacy PHP applications. Managing technical debt, incremental improvements, and maintaining team sanity during large-scale code transformations.

Distributed Monolith - How NOT to do Microservices

Anti-patterns in microservices adoption. How organizations accidentally create distributed monoliths and the hidden costs of premature decomposition.