Umarfarook GurramkondaResume (PDF, opens in a new tab)Book a call (opens in a new tab)

Umarfarook Gurramkonda

I build LLM features and the evals that check them. ML engineer at Hypeon AI since Oct 2024. My public projects, with their test results, are on GitHub and PyPI.

Book 15 minutes (opens in a new tab)Resume (PDF) (opens in a new tab)

Open to full-time roles and contract work, remote or in Bengaluru.

GitHub (opens in a new tab)LinkedIn (opens in a new tab)

Illustrated avatar of Umarfarook Gurramkonda
  • MCP
  • Pydantic
  • Claude
  • Gemini
  • OpenAI
  • Groq
  • Vertex AI
  • AWS Bedrock
  • LangChain
  • Mastra
  • Vercel AI SDK
  • Hugging Face
  • PyTorch
  • scikit-learn
  • pandas
  • YOLOv8
  • OpenCV
  • Python
  • FastAPI
  • Flask
  • Django
  • SQLAlchemy
  • PostgreSQL
  • Drizzle
  • Redis
  • BigQuery
  • Node.js
  • TypeScript
  • Next.js (React)
  • Tailwind CSS
  • Zod
  • AWS ECS Fargate
  • GCP Cloud Run
  • Docker
  • GitHub Actions
  • Prometheus
  • Linux
  • Stripe
  • PostHog
  • Meta Marketing API
  • pytest
  • ruff
  • Postman
  • Claude Code
  • Cursor

Expertise

I am a machine learning engineer in Bengaluru with 2+ years of work on LLM systems. At Hypeon AI I work on the agent, retrieval and generation features behind its ad research and ad creation products. On my own time I publish small, tested projects: an MCP server on PyPI, an eval harness for support agents, and a language model written from scratch.

  1. LLM agents

    Multi-step agents that stream their answers and return structured output.

    • Claude
    • Gemini
    • OpenAI
    • LangChain
    • Pydantic

  2. Retrieval and RAG

    Retrieval with LangChain and FAISS, with guardrails against made-up answers.

    • LangChain
    • FAISS
    • Hugging Face
    • sentence-transformers

  3. NL-to-SQL and cost guardrails

    Plain-English questions become SQL, and each query's cost is checked by a dry run before it runs.

    • BigQuery
    • MCP
    • PostgreSQL

  4. Evaluation

    Golden sets, deterministic checks and a temperature-0 judge, gated in CI.

    • pytest
    • GitHub Actions

  5. Models, trained and from scratch

    A transformer, a diffusion model and a plate detector, built in PyTorch and reported with the numbers from their own runs.

    • PyTorch
    • scikit-learn
    • YOLOv8
    • OpenCV

  6. Full-stack delivery

    The API, the web app and the deploy around the model.

    • FastAPI
    • Next.js
    • Docker
    • AWS
    • GCP

Stack

LLM systems
  • Agents
  • Tool calling
  • RAG
  • MCP
  • Context engineering
  • Pydantic
  • SSE streaming
  • LLM-as-judge evals
  • Fine-tuning smaller models
  • Voice AI (TTS/STT)
Models and frameworks
  • Claude
  • Gemini
  • OpenAI
  • Groq
  • Vertex AI
  • AWS Bedrock
  • LangChain
  • Mastra
  • Vercel AI SDK
  • Hugging Face
  • sentence-transformers
  • FAISS
ML and vision
  • PyTorch
  • scikit-learn
  • XGBoost
  • pandas
  • YOLOv8
  • OpenCV
  • EasyOCR
Backend and data
  • Python
  • SQL
  • FastAPI
  • Flask
  • Django
  • SQLAlchemy
  • PostgreSQL
  • Drizzle
  • Redis
  • BigQuery
  • Node.js
Web
  • TypeScript
  • Next.js (React)
  • Tailwind CSS
  • Zod
  • Phaser 3
Cloud and delivery
  • AWS ECS Fargate
  • AWS S3
  • AWS EC2
  • AWS RDS
  • AWS SES
  • GCP Cloud Run
  • Vercel
  • Docker
  • GitHub Actions
  • Prometheus
  • Linux
Product APIs
  • Stripe
  • PostHog
  • Meta Marketing API
Testing and tooling
  • pytest
  • mypy
  • ruff
  • Postman
  • Claude Code
  • Cursor
Familiar
  • LangGraph
  • LangSmith
  • CrewAI
  • AutoGen
  • DSPy
  • TensorFlow
  • Kubernetes
  • Kafka
  • Snowflake
  • Databricks
  • gRPC
  • Azure

Experience

2+ years across two roles.

Resume (PDF) (opens in a new tab)

  1. Founding ML Engineer

    Hypeon AI, Bengaluru, remote

    Oct 2024 to now2 yrs

    Hypeon AI makes an AI ad platform for D2C brands: ad research and an ad creation studio. The company was in stealth until Sep 2025.

    • Built LLM agent features: research agents that stream their answers and return structured output.
    • Built NL-to-SQL over the ad data, with each query's cost checked before it runs.
    • Built image and video generation for HypeOn Studio, with fallbacks across model providers.
    • Work with customers on requirements, demos and rollout.
    • Python
    • FastAPI
    • PostgreSQL
    • Claude
    • Gemini
    • OpenAI
    • AWS
    • GCP

    The code is private to the company. I can talk about my part of it in an interview.

  2. Backend Developer Intern

    Synclovis Systems, Bengaluru

    Jun to Sep 20244 mo

    • Built REST APIs in Node.js, Express and MySQL for an event-management web app, with Redis caching.
    • Contributed LangChain and FAISS retrieval to an internal RAG healthcare chatbot.
    • Node.js
    • Express
    • MySQL
    • Redis
    • LangChain
    • FAISS

Work

Public projects, each with the numbers from its own runs and a link where you can check them.

All work (9)

Check the work

Have a role or a project? Let’s talk.

Hiring for an AI or ML role, or need an LLM feature built and measured? Write to me.

Open to full-time roles and contract work, remote or in Bengaluru.

Email meBook 15 minutes (opens in a new tab)

How a project starts

  1. 15-minute call

    You show me the product or the feature. I ask where it fails and what it costs.

  2. Written scope in 24 hours

    One page: what I will deliver, when, for how much. You reply yes or no.

  3. Work starts within a week

    Daily written updates. You see working software before you see a slide.