What a Forward Deployed Engineer Actually Does
Last week I started as a Forward Deployed Engineer intern at CallKaro AI in Gurgaon. CallKaro builds voice agents that pick up customer calls and WhatsApp chats in 20+ languages. When I told people the title, most of them asked what it meant. Fair question. I had to look it up too before the interview.
The short version: a normal engineer builds the product, an FDE makes the product work for one specific customer. You sit between the client and the platform. You join the discovery call, figure out what their calls actually sound like, and own the setup until the agent is live and handling real traffic.
A lot of the job is prompt work. A voice agent is mostly a persona plus a call flow, and both live in the prompt. Getting it to sound natural in Hindi or Hinglish is its own problem, because people switch languages mid-sentence and the agent has to keep up without getting weird about it.
The other part is debugging live deployments. When an agent says something wrong on a call, there is no stack trace. You read the call log and the transcript, find the turn where it went off track, and work out whether it was the prompt, the transcription, or the flow. Then you fix it and watch the next batch of calls.
And then there is plain integration work: Python scripts and API calls that connect the platform to whatever CRM or workflow the client already uses. Not glamorous, but the agent is useless if the lead it just qualified never shows up in their system.
Coming from embedded AI at STMicroelectronics, the change of pace is the biggest thing. Firmware bugs took days to reproduce. Here a bad prompt shows up in the next hour of calls. I'll write more once I've shipped a few deployments end to end.