Applied AI Systems Engineer

Gurgaon

Looking for someone who can turn “Disha forgot again” into an actual working memory system.
You prototype fast, evaluate harder, and enjoy grinding 80% into 99% without needing a new shiny toy every Tuesday. You’ve built real LLM stuff (prompts, tools, RAG, memory) and can explain what broke in production.
Bonus points if you have strong opinions on the latest and greatest of the AI world like which is the best harness and aren't afraid to debate it out.

Full JD - Link

Name *

Email *

LinkedIn profile *

Github *

Phone *

Describe an AI system (or a clear part of one) you personally built. *

Problem you were solving and how did you solve it? Wispr flow dictations preferred.

Did this ship to real users? *

Tell us about a long-tail failure you found after it was live. *

How did you notice it (logs, traces, user complaints, evals, reviews)? What did the iteration look like: hypothesis → change → how you checked it actually got better?

Which model would you use for Disha’s live voice conversations, and with which provider? *

Assume real-time voice in Hinglish with mid-call tool use, hundreds of thousands of users, tight latency (about 1–2s to respond), healthcare safety stakes, and real cost pressure. Pick a model and provider, then briefly say why that combo over the obvious alternatives on quality, latency, tool calling, cost, and reliability.

Made withBinary