Nilesh Khedkar.
← All work

Expert-built AI agent platform for maintenance

A multi-agent platform where maintenance and reliability experts build, deploy and supervise their own AI agents without writing code.

Domain
Enterprise maintenance
My role
Product owner and implementation lead
When
2024–2026
Status
In production
Result
<1 hr

to build and ship an agent, down from 2–3 weeks

  • Subject-matter experts build and deploy an agent in under an hour, against a previous two-to-three-week build cycle
  • Specialist agents share one retrieval layer and a persistent knowledge-graph memory

The problem

Experts in maintenance and reliability spend most of their time on legwork: gathering data from SAP and document stores, chasing people for context, and writing up findings. Only a fraction of their day goes to the judgment they were hired for. Central IT teams could not build AI tools fast enough to change that, and the tools they did build rarely matched how the work is really done.

The approach

Flip the model. The expert is the best person to automate their own work, so the platform lets them do it: they design the data flow, write the instructions, test against their own cases and supervise the agent the way they would coach a new engineer.

Under the hood, the platform follows a harness-first design. Deterministic scripts handle retrieval, calculations and document generation; the language model is used for the parts that need judgment. Each agent has a defined role, responsibility and human owner.

What changed

  • Agent build time fell from weeks to under an hour.
  • Experts now own their agents end to end, including their quality.
  • The same pattern is being extended to voice-first assistants for field technicians.

Say hello

Always happy to swap notes on AI in industry, maintenance and reliability.