06 From the workshop
Across engagements the same infrastructure gaps appear: nobody can prove a model is still good, nobody can constrain an agent, nobody can update software on a segmented site. We are hardening our answers into products.
01 In the workshop
Each one exists because its absence has cost a client real time. Listed with the failure it removes.
T1
In development
Slice-aware offline and online evaluation for industrial models and agents, with drift monitors, shadow scoring and promotion gates wired into CI. Built because every engagement needed it and no two teams had it.
Removes Nobody can answer "how would we know if this got worse?"
T2
In development
Tool contracts, sandboxed execution, authorisation boundaries, trace capture and replay for agentic systems — the scaffolding that makes an agent auditable rather than merely impressive.
Removes Agents that cannot be reviewed, reproduced or constrained.
T3
Prototype
Versioned model rollout, shadow evaluation and rollback across gateways on constrained, intermittently connected, segmented sites.
Removes Model updates that require someone to visit the plant.
T4
Research
Lineage capture from instrument to published figure, designed to satisfy a reproducibility audit rather than to populate a dashboard.
Removes A result nobody can regenerate eighteen months later.
02 Also from the lab
Not our business, but a useful discipline — and where the company name came from.
SkyWiki is an astronomy application we build and maintain in-house — on-device celestial mechanics, sensor-fused augmented reality and satellite pass prediction, computed offline on a battery budget.
We keep it shipping because the constraints rehearse the ones that matter in industrial work: numerical accuracy that has to hold at the edges, sensor fusion against noisy reality, hard performance envelopes, and maintaining a system across years of platform change. A team that has never carried that has never been properly tested.
Early access
The tooling ships first to organisations we are engaged with, and occasionally to teams whose problem is sharp enough to be worth designing against. Tell us which gap you are trying to close.
or write directly — hello@bluestreaklabs.com