AI Agent Platform Case study

Agent.Press

The agent platform the engagement is built on: an LLM agent framework with an admin console, an extensible tool system and retrieval pipelines for running production assistants.

Role
Senior Full-Stack AI Engineer · Architecture & Platform
Disclosure
Public product · Proprietary implementation omitted

Verified proof

One agent runtime carrying several production client systems

Agent.Press title card
Approved, public-safe project imagery.

The problem

What the system had to solve.

Every engagement needed the same foundation underneath it: retrieval, tool execution, conversation state and a console to configure and watch an agent from. Rebuilding that per client is where reliability goes to die.

What I owned

The responsibility behind the work.

Architected platform capabilities as a shared foundation and delivered client systems on top of them, so the same agent runtime carries several production products.

The Sage Intacct automation for Reis Contracting ships against the Agent.Press API, library and interface packages, so platform work and product work moved together rather than drifting apart.

Engineering approach

How it was built.

Worked across the agent runtime and its surfaces: configuration and conversation state, an extensible tool system, retrieval pipelines over a vector-enabled Postgres store, and the console and drop-in chat clients above the API.

Agent request path

  1. Request
  2. Context
  3. Tools
  4. Validation
  5. Result

Engineering scope

What I worked on.

  1. Agent configuration, tool execution and conversation state
  2. Retrieval pipelines over a vector-enabled Postgres store
  3. Console and drop-in chat surfaces above the API
  4. Client systems delivered on the platform
  5. Backend services, third-party integrations and release flow

Impact

The verified result.

Client systems ship against the platform packages rather than a private fork, so the ERP automation, the reporting workflows and the internal assistant share one runtime, one tool contract and one upgrade path.

Technology

The technical surface.

  • TypeScript
  • Next.js
  • PostgreSQL
  • Vector retrieval
  • OpenAI
  • Tool calling

Next: Local Falcon

Back to all work Start a conversation