Jagdish Bhatt

I’m an AI engineer focused on building production-grade LLM agent systems and end-to-end ML pipelines. I ship complete systems (not just notebooks), with a strong bias toward reliability, latency, and measurable outcomes.

In my Nexarch live project, I built a LangGraph-powered multi-agent orchestration system using Azure OpenAI and Gemini to reason over 10K+ live microservice traces/day. I reconstructed production architectures into real-time dependency graphs without source-code access, and used prompt engineering plus tool-augmented agents to automate bottleneck classification and architecture comparison.

I also emphasize production observability and optimization. With OpenTelemetry, I created telemetry capture with <5% runtime overhead, and implemented an LLM optimization engine that generates architecture variants and recommends the best one—reducing manual system analysis effort by ~60% while cutting inference latency by ~60%.

Earlier, I built Pindora Shield, an AI drug discovery framework that scales to 10K+ molecules per run using GANs (TenGAN) and multi-model ML pipelines. I automated a full 6-stage workflow (disease input to target mapping, molecule generation, multi-model property prediction, filtering, and ranking) with failure-safe execution—reducing candidate screening time by >90%—and I’m excited to apply this same “end-to-end” mindset to financial research and market intelligence.