General questions
Most of our projects run between six and twelve weeks from the initial data audit to a deployed, monitored system. Smaller automations, like a single document-extraction pipeline, can be live in three to four weeks. Larger multi-model platforms with custom dashboards have taken up to five months, though that is the exception rather than the rule.
Not necessarily. We can work directly on your infrastructure via a secure VPN or jump box, so the data never leaves your environment. If remote access is not possible, we can use anonymised or synthetic samples for initial prototyping and then fine-tune on the real data once we are on-site. We sign a data-processing agreement before any engagement begins.
Our strongest track record is in retail, logistics, property management and financial services. That said, the techniques we use, including time-series forecasting, NLP and computer vision, transfer well across sectors. We recently completed a defect-detection system for a food-packaging manufacturer and a patient-triage chatbot for a private clinic group. If your problem involves structured or semi-structured data, we can almost certainly help.
It depends on the task. A demand-forecasting model for a retailer with 50 SKUs and two years of daily sales data is perfectly feasible. A computer-vision model for defect detection might need 2,000 to 5,000 labelled images. During the discovery phase we assess your data volume, quality and labelling status, and we will tell you honestly if you need to collect more before a model can deliver useful accuracy.
Yes. We deploy models as REST APIs or containerised microservices that slot into most modern architectures. We have integrated with Salesforce, SAP, Shopify, various ERP systems and bespoke internal platforms. If your stack uses an older protocol or file-based data exchange, we build an adapter layer so the model fits without forcing you to re-architect everything else.
Commercial and support questions
Every project includes twelve months of monitoring and maintenance. We track model accuracy, data-drift metrics and inference latency through a shared dashboard. If performance drops below the threshold we agreed on during scoping, we retrain the model at no additional cost. After the first year, you can extend support on a monthly retainer or bring maintenance in-house using the documentation and runbooks we provide.
Both. For well-defined projects where the scope is clear, we quote a fixed price. When the scope is exploratory, for example a research sprint to determine whether a particular approach is feasible, we work on a weekly time-and-materials basis with a cap you set in advance. We never bill beyond the agreed cap without written approval.
You do. All custom code, trained model weights, documentation and deployment scripts produced during an engagement belong to you upon final payment. We retain the right to use general techniques and open-source libraries we contributed to before the project, but nothing specific to your data or business logic.
We define success metrics during the scoping phase, before any contract is signed. If after two full iteration cycles the model still falls short, we write up our findings, explain why and what additional data or approach changes could close the gap, and you pay only for the work completed up to that point. We do not charge for a deliverable that does not meet the specification.
Security is non-negotiable. All data in transit is encrypted with TLS 1.3, and data at rest sits on AES-256-encrypted volumes. Our team members access client environments through individual, auditable accounts with multi-factor authentication. We are happy to undergo your internal security review or provide evidence of our own penetration-testing results before work begins.