We build production software on the systems you already run.

The model is the easy part. The records are the hard part. Years of Zendesk tickets where the useful answer is in the third reply of a merged thread. Call recordings where the account number is said twice, differently. A product catalogue with four SKUs for one product. Internal documentation that was last true before the reorganisation. We build the retrieval layer over that: chunking that respects where a document actually ends, an embedding index and a keyword index because neither is enough alone, metadata filters so a question about the current contract is not answered from the one it replaced. Every answer cites the passage it came from. When retrieval finds nothing good, the agent says so, in words, and hands off to a person.

A confident answer from the wrong document is the failure that reaches a customer with your name on it. It is the one we design against first, and the one the evaluation set is built to catch. If your records cannot support what you want to ask of them, we say so at the scoping stage, before anyone has paid for the attempt.

You run the marketplace. You hold the licence, the merchant account and the contract with the payment provider. We are the engineers you hire to build what sits behind it. Catalogue and search, including the part where two sellers list the same product under different titles. Seller onboarding and identity checks, wired to the verification provider you chose. Order state, so a cancellation after dispatch does not leave a payout half made. Payout calculation, which your own licensed provider settles from your accounts. And the operations tooling nobody demos: the refund queue, the dispute screen, the seller suspension button, and the reconciliation report that says what your ledger thinks the provider owes and what the provider says it paid. Every marketplace needs that report.

Money moves through your accounts at your licensed provider. We do not operate marketplaces, and we do not hold, receive or transmit funds for anyone. We write the code that instructs your provider, and the code that checks the instruction was carried out.

Your Zendesk, your Salesforce, your Shopify, your call recordings, your internal APIs. Under your credentials, with your written authorisation, through the interfaces those vendors document, with the narrowest permissions that do the job. Getting the data out once is a script. Keeping it flowing is the work. Shopify meters you with a leaky bucket. Salesforce gives you a daily quota and an error for everything past it. Zendesk answers with a 429 and a Retry-After header that has to be honoured. Pagination that changes shape halfway through a backfill. A field that was a string in March and an object in April, with nothing in the changelog. Deleted records that never appear in an incremental pull unless you ask for them by name.

We build the pipeline that handles all of that, lands the data in a warehouse you own with a schema you can read, backfills history without tripping the limits, and alerts a person when the shape of the data drifts.

Every automated workflow is a promise that a person no longer needs to look. Lead arrives, gets enriched, gets routed, gets a first reply. Ticket arrives, gets classified, gets the right macro or the right human. Stock drops below threshold, purchase order drafts itself. Email sequence fires on a real event instead of a timer. None of this is hard to build. It is hard to build so that it fails well. What happens when the enrichment provider is down for an hour. What happens when the ticket is actually three tickets in one message. What happens when the stock count is wrong because a return was booked twice.

We write those paths first, then the happy path. Every automated action is logged with what it saw and why it decided. Every one can be replayed. Every one has an off switch a non-engineer can reach.