AI that works. Not just on demo day.
I’m Jan Wachowiak. Through my company, Neural Networks, I build AI agents, assistants on company knowledge and models that run on your own servers.
Since 2021 I have built AI for Fortune Global 500 companies, a Polish government agency and teams in finance, legal tech, health tech and e-learning.
Scroll
Most AI is hype. I build the opposite.
Stable systems that solve real problems. And keep solving them after launch.
A demo is the easy part. Production is where it gets hard:
- Real data.
- Real users.
- Real costs.
I take systems from prototype to production, with quality tests, safeguards, monitoring, and cost and speed tuning.
See what I buildAI that takes work off your team.
Three kinds of systems. Each one built for production.
-
Agents that do the work.
Agents handle repetitive steps in your processes on their own: documents and contracts, quoting, ticket handling, reporting. A human stays in control wherever it matters.
-
Answers from your own documents.
Assistants and search that answer employees and customers from your documents, procedures and databases. Every answer shows its source.
-
Your data stays with you.
Language models on your own infrastructure, with nothing sent outside. I choose the model and hardware, tune performance, and you keep full control of your data and GDPR compliance.
Where AI pays off.
I assess where AI makes business sense, choose the approach (cloud, local or mixed), and plan the rollout with costs and risks on the table.
Tested before it talks to anyone.
Answer quality tests, safeguards against errors and abuse, monitoring and cost control after launch.
Large companies trust me with their AI.
Fortune Global 500 companies in energy and advanced manufacturing. Finance and sales at a large corporation. A Polish government agency.
Clients and partners: EPAM Systems, TechTorch, Transition Technologies MS, Metric Space, IMIN, and 10+ companies through Upwork.
Industries: energy, manufacturing, finance, legal tech, health tech, e-learning, public sector.
End client names are under NDA.
See the workReal systems. Real constraints.
A few things I have built. Client names stay under NDA where they have to.
-
Fortune Global 500energy and advanced manufacturing
Agents and enterprise RAG
Agent systems and enterprise RAG platforms on large-scale projects. I own the architecture, set the technical direction, review the team’s work and take solutions from prototype to production.
-
A large corporationwith TechTorch
Agents for finance and sales
Agents that automate internal finance and sales work: quoting in Salesforce CPQ, contracts in Conga CLM, analysis of large document sets. I own the agent architecture and integrations.
-
A Polish government agencyname under NDA
A 12B model on the agency’s own servers
An agentic RAG assistant for a citizen-facing portal, built on the agency’s own data. A 12B open-weight model set up entirely on the agency’s hardware: serving, quantisation and inference tuning under strict data sovereignty rules.
-
Transition Technologies MSAI4Legal, AI4Elearning, AI4Rental, AML Track
AI for legal, e-learning, rental and AML
Production RAG systems, high-throughput FastAPI services and agent architectures on Azure. Backend design, data flows, deployments and reliability under heavy load.
-
Metric Spaceindustrial AI
Agents inside secured production environments
LLM agents with function calling for manufacturing clients, running on their own in secured environments. Fine-tuned models, GDPR anonymisation models and benchmarks for retrieval quality and agent reliability.
-
IMINhealth tech
Breathing matched to your nervous system
A system that reads the state of a user’s nervous system from physiological signals (rPPG) and recommends breathing techniques. Production inference on AWS.
-
Upwork
15+ projects for 10+ companies
LLMs, agents, NLP, computer vision and classical ML. Top Rated Plus (top 3% of freelancers) and a 100% Job Success Score.
From first conversation to production.
I do the work myself, end to end.
Analysis
Where AI brings a real gain.
Architecture
Cloud, local or mixed. Plan, costs, risks.
Build
Agents build inside my harnesses, I set the direction.
Launch
Go-live and team training.
Growth
Monitoring, quality and costs under control.
I decide what ships
I work AI-native. Agents do a large part of the building, inside harnesses I design for development, testing and evaluation. I set the direction and decide what ships. It is why I can deliver at the pace I do.
What I work with LLM agents and multi-agent orchestration, RAG, fine-tuning and self-hosted models, Python and FastAPI, Azure, AWS and GCP, evaluation and monitoring.
The person behind Neural Networks.
Jan Wachowiak. Founder and Lead AI Engineer.
I started Neural Networks in 2021. Everything I have built since has gone through it. I learned AI by shipping real projects. I work remotely from Toruń, Poland, in Polish and English. Agents are what I know best.
LinkedIn profileA process that eats your team’s time?
Or data you can’t send to the cloud? Write to me.
Describe the process or the problem in a few sentences. We will start with a conversation about whether AI makes sense there.