Software and AI projects
Solutions for
specific problems.
Eight common problems: what we would build, what your team receives and how we check the result.
Illustrative solution briefs. Scope and measures are agreed for each engagement.
Build custom software, reviewed by your team.
A custom product needs design, architecture, tests and someone to maintain it. Coding agents speed the work up, but their code has to be reviewed like anyone else’s.
Related expertiseThe approach
We build the product in releases you can try. Agents investigate issues, propose changes and run tests in an isolated copy of the repository. Your team reviews the code and decides on deployment.
What we deliver
- Product, architecture and agent task boundaries
- Code changes with tests and execution evidence
- Review, automated delivery and team documentation
What we measure
Changes accepted after review, detected regressions, correction effort and time from issue to delivery.
Prepare proposals from your own documents.
Preparing a proposal means gathering scattered requirements, services and references. Copying earlier proposals can carry forward commitments that no longer apply.
Related expertiseThe approach
We connect the catalogue and approved references. The assistant checks requirements against those sources and prepares the draft; the team reviews its commitments before sending it.
What we deliver
- Sources, permissions and freshness criteria
- Proposal assistant with references and review
- Evaluation cases for coverage and unsupported claims
What we measure
Requirement coverage, reference accuracy, unsupported commitments and review time.
Resolve incidents across systems.
A failed delivery requires checking orders, transport and warehouse systems. Decisions and changes are spread across tools and people.
Related expertiseThe approach
We connect the lookups and actions the agent needs, with permissions for each tool. The team decides which operations it can carry out and which need approval. If data is missing or a lookup fails, the case stays open with the completed work saved.
What we deliver
- Resolution workflow and authority boundaries
- Connections to orders, warehouse and delivery systems
- Approvals, action records and failure recovery
What we measure
Completed resolutions, incorrect or duplicate actions, manual interventions and time to resolution.
Reconcile invoices, orders and deliveries.
Extracting an invoice does not confirm that billed items match what was ordered and received. Your team needs to compare documents and resolve differences.
Related expertiseThe approach
We read and link purchase documents, then apply the agreed reconciliation rules. Each discrepancy keeps a reference to the original documents so its owner can check it.
What we deliver
- Reading and linking invoices, purchase orders and delivery notes
- Reconciliation rules and a discrepancy queue
- Purchasing system integration and decision records
What we measure
Field accuracy, correctly linked documents, detected discrepancies and false alarms, resolution time.
Test AI changes before release.
Changing a model or its instructions can alter an agent’s answers and actions. The team needs to detect those changes before they affect users.
Related expertiseThe approach
We prepare test cases and evaluation criteria with the team. We check answers and changes in connected systems, repeat runs and review automated scores. Production failures become cases for subsequent tests.
What we deliver
- Evaluation datasets and scoring rubrics
- Automated scoring checked against reviewers, and regression tests
- Release criteria and quality reporting
What we measure
Task success, judge agreement with reviewers, regression rate, failure coverage and quality by user segment.
Share model access across products.
Separate experiments accumulate different provider integrations, prompts and monitoring. Costs and behaviour become harder to understand as adoption grows.
Related expertiseThe approach
We build shared model access with spending limits and fallbacks. Each request records a trace and the version of its instructions. We assess routing and caching against the task, permissions and maintenance cost.
What we deliver
- Model gateway and service interfaces
- Tracing, cost controls and dashboards
- Deployment and operational runbooks
What we measure
Routing correctness, cost per completed task, p95 latency, availability, incorrect cache reuse and time to release a change.
Move an AI prototype into production.
A prototype may work with one user and a small dataset, yet still lack the deployment controls, capacity planning and data boundaries required for everyday use.
Related expertiseThe approach
We design deployment on AWS, Google Cloud or Azure around the workload and existing systems. We compare managed services with private or container hosting. We automate environments and prepare identity, monitoring and recovery.
What we deliver
- Cloud architecture and service decisions
- Infrastructure as code and deployment pipelines
- Capacity, cost and operational runbooks
What we measure
Availability, recovery time, deployment reliability, infrastructure cost and performance under representative load.
Forecast demand and prioritise cases.
Teams need to prioritise work, anticipate demand or identify unusual behaviour. Reports describe what happened, but provide limited help with the next decision.
Related expertiseThe approach
We agree which decision the model will support and the cost of getting it wrong. We compare predictions with a simple baseline using held-out data and realistic evaluation periods. Then we bring them into the workflow with monitoring and human review.
What we deliver
- Data assessment and benchmark baseline
- Prediction service and workflow integration
- Performance monitoring and retraining criteria
What we measure
Forecast error, precision and recall, calibration, adoption and the effect on the decision the model supports.
Not sure where
to begin?
Tell us which part of the work needs improving.
We look at the data and tools you already have. From there we can tell you whether to build something custom, buy an existing product or run a small trial first.
Explore strategy