AI Agent Development Company

Description

We are a mid-size logistics operator and we have been burned once already by an agency that demoed something impressive and left us with a prototype we could not run. This time we want a development partner rather than a build-and-vanish vendor. The scope is a small fleet of agents that sit over our TMS: one that reads carrier emails and updates shipment status, one that drafts exception reports for the ops desk, and one that answers "where is my order" from our own data. What matters to us is that the second agent can be handed to our internal team six months from now — so we care about how you structure the code, how the prompts are versioned, and what happens when a model provider changes behaviour under us. Tell us in your proposal about a project you handed over, not one you are still running.

• Design the agent architecture as separable services, so one agent can be replaced without touching the others
• Version prompts and tool definitions in the repository, not in a vendor console we cannot export
• Build the carrier-email agent against our real inbox format, including the three carriers that send HTML tables
• Set up evaluation cases we can re-run after any model or prompt change, with pass/fail we can read
• Run two handover sessions with our internal developer and leave the recording

• Three agents deployed to our environment, running against production data
• Repository with prompt versioning, evaluation suite and CI
• Architecture note explaining what to change when a provider deprecates a model
• Handover sessions recorded, plus a written runbook