FAQ
Pricing, how engagements run, who owns what afterwards, and the answers we give before anyone signs anything.
Most work is fixed-scope for a defined phase rather than open-ended time and materials. We scope in about a week, give a cost envelope before any build starts, and break larger programmes into phases you can stop between. For exploratory work we use a time-boxed advisory sprint.
One workflow, in production, in six to ten weeks — roughly a week agreeing the success measure, two on retrieval and the evaluation harness, then build and hardening. Narrow enough that you can judge it honestly and cancel cheaply if the economics do not hold.
Yes, and that is the common case. We pair with your engineers rather than delivering over a wall, and handover documentation and runbooks are part of the deliverable rather than an upsell.
Whichever measures best for the workload, behind an abstraction that lets you switch. We route per task on cost, latency, and accuracy across Claude, GPT, Gemini, and open-weight models. We hold no reseller relationships, so the recommendation is made on measurements alone.
You do. Code, prompts, test sets, and the evaluation harness live in your repository. If you later move to another team, your quality bar moves with you — we do not treat that as a lock-in mechanism.
Dubai, Greater Seattle, and West Bengal, India — a follow-the-sun model across three continents. In practice what matters is overlap hours with your team, which we agree at the start of an engagement.
Yes. We routinely work under client paper, sign NDAs before detailed scoping, and can accommodate security review as part of onboarding.
We will say so. We have recommended buying an off-the-shelf product and fixing a data model instead of building, and expect to again. An assessment that removes something from your roadmap has done its job.
Still deciding
The questions we would ask if we were the ones buying.
Assessment and feasibility before committing to a build.
Tell us what you are trying to build.
