Capabilities
Four kinds of system, not a list of technologies.
Nobody buys machine learning. These are the systems people actually ask for, what is genuinely hard about each, and the work that shows we have built it.
Search that understands what someone meant, over inventory or documents that were never organised for searching — and answers that carry the passage they came from, so a reader can check rather than trust.
- Search engines over inventory, catalogues and listings
- Enterprise knowledge systems across mixed document formats
- Retrieval pipelines that resolve loose phrasing into real constraints
- Grounded assistants that answer only from what was retrieved
Proven in: SolvTrav, AskQuo
Operational systems that carry the routine work and escalate the rest — where the interesting design question is not what the agent can do, but what it must never do without a person.
- Workflow automation across systems that were never integrated
- Operational assistants and daily decision boards
- Business process automation with explicit exception handling
- Human-in-the-loop steps at the points that carry risk
Proven in: Recruitment Operations
Visual systems that generate against a real catalogue rather than against the model's imagination — because a beautiful sofa the retailer does not stock is not a feature, it is a support ticket.
- Visual commerce: real products placed in a customer's own space
- Controlled generation pipelines constrained to real inventory
- Multimodal systems that read an image and act on what is in it
- Revision by description rather than by editor
Proven in: AI Space Visualizer
The unglamorous half: the backend, the data path, the deployment, the controls and the measurement. Most AI work that fails does not fail at the model — it fails here.
- Backend architecture and AI APIs built for real traffic
- Data pipelines feeding retrieval and generation
- Deployment, gating and rollback that a team can operate
- Monitoring, spend controls and the switches to turn a feature off
Proven in: AI Space Visualizer, SolvTrav
How we build
AI that works inside real systems.
Grounded in business data
AI does not invent company truth. Answers come from what was retrieved, and they arrive with the source attached.
Human fallback on the paths that matter
Which decisions require a person is decided at design time and enforced in the workflow, not asked for in a prompt.
System truth is not model output
The application owns business state. A model reads it and proposes changes to it; it does not author it.
Measured in workflows
Success is work getting done, with fewer interventions, at a known cost — not a score on a model benchmark.