Working with
gigantic context,
without losing it.
I help deeptech teams build durable context systems — multi-GB documentation, enterprise codebases, and agent pipelines where knowledge compounds instead of decaying.
Context Engineering at Scale
Multi-GB documentation, enterprise codebases, 6+ analyst teams. Retrieval, indexing, and abstraction layers that keep knowledge alive instead of lost.
Forward Deployed AI
Shipping agentic systems inside real organizations - not demos. Realtime agents, tool use, custom LSPs, domain-tuned models.
Continuous Post-Training
LoRA and cyclic post-training pipelines that turn your own corpus into a serving layer. Models trained on your context, refreshed on your cadence.
Local-First Inference
On-device and on-prem model gateways. Privacy, latency, and cost under your control. Author of ppmlx, an open-source MLX gateway for coding agents.
Rafał is the rare engineer who can hold an entire enterprise architecture in his head - and then teach a system to hold it too. He took our multi-gigabyte documentation estate, the kind of thing six analysts drown in, and turned it into living context that our agents actually reason over. Every spec traces to a module, every module to working code. Nothing gets lost anymore.
Most consultants demo well and disappear at deployment. Rafał is the opposite - he ships inside your org until the thing works without him. Realtime agents with tools, domain-tuned models, retrieval layers tuned to how your team actually thinks. If you are serious about forward-deployed AI, this is who you want in the room.