Skip to content
View subratamondal1's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report subratamondal1

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
subratamondal1/README.md

Subrata Mondal

AI Engineer · agentic AI · applied LLMs · production backends

Python · LangChain · LangGraph · MCP · multi-agent · evaluation

Portfolio · LinkedIn · Email


What I work on

Production agentic AI at lawworld.ai. Multi-agent pipelines, LLM-powered backends, evaluation harnesses, and the ops infra to keep it all running. Multiple Python services covering agent architectures, prompt + context engineering, eval, and deployment at scale.


Stack

  • Language — Python 3.12+ · async · type-annotated end-to-end
  • Web — FastAPI · Pydantic · httpx · structlog
  • Agents + LLMs — LangChain · LangGraph · ReAct · MCP · structured outputs · LLM-as-judge · prompt caching
  • Data — MongoDB Atlas (async) + Atlas Vector Search · native Voyage AI embeddings & reranking · Redis
  • Cloud — Azure (Container Apps · Service Bus · Blob · Key Vault · AI Foundry) · Docker
  • Infra + Ops — Bicep · Terraform · GitHub Actions · OpenTelemetry · ARQ
  • Tooling — uv · ruff · ty · pytest · orjson

Strengths

  • Agentic systems — multi-agent orchestration · ReAct loops · planning · reflection · tool-calling · structured outputs
  • LLM reliability — retries · fallback · provider routing · prompt caching · cost + latency engineering
  • Evaluation — golden datasets · LLM-as-judge graders · eval harness · metrics-driven iteration
  • Backend depth — async Python · schema-first APIs · queue-based pipelines · observability · IaC
  • Durable orchestration — long-running agent runtimes · event-driven choreography over message queues · per-step checkpointed crash-recovery · self-healing producer-worker pipelines · idempotent at-least-once delivery
  • Production ownership — incident response · on-call · deployment · cost attribution

Currently exploring

  • MCP servers · A2A protocol · agent-tool ecosystems
  • Kubernetes-native deployment patterns (production currently on Azure Container Apps)

At a glance

role: AI Engineer
focus: agentic systems · applied LLMs · production backends
stack: Python · async · FastAPI · LangChain · LangGraph · MCP · MongoDB · Azure
interests: hard agent / LLM problems · contract · open-source collaboration · technical speaking
timezone: IST (UTC+5:30) · async-first remote work
contact: [email protected]
linkedin: https://www.linkedin.com/in/i-am-subrata-mondal/
portfolio: https://subrata-mondal-portfolio.netlify.app/

Pinned Loading

  1. intelligent-ocr intelligent-ocr Public

    Effectively uses multiple steps through Azure Vision, GPT-4o and Claude to get highly accurate instructions.

    Jupyter Notebook 2