My name is Sameer Shanbhag — a software engineer who's happiest building AI agents, the infrastructure around them, and full-stack things that actually ship.

Here is the stack I like to work with:

And the AI toolchain I spend most of my time in:

To know more about me please keep reading...

• 📍 Location: Sunnyvale, California (Bay Area)
• ✉️ Email: sameershanbhag14@gmail.com
• 💼 LinkedIn: LinkedIn/sameershanbhag
• 🌐 Portfolio: sameershanbhag.com
• ✍️ Blog: sameershanbhag.com/blogs

These days I'm a Staff Software Engineer at Walmart Global Tech, where I get to work on agentic systems (including Marty, one of Walmart's AI assistants) and, before that, the billing side of its advertising business. The road here ran through audio tooling at Qualcomm, bioinformatics research during my M.S. in Computer Science at UNC Charlotte, and fintech at Morgan Stanley — a winding path that turned out to be great preparation for agents, where distributed systems, data, and product all collide. I like the unglamorous parts of the craft: state, recovery, observability, and making things reliable enough that people can depend on them.

• Languages: Python, Java, TypeScript, JavaScript, HTML5, CSS3, SQL
• AI & Agents: LangGraph, LangChain, checkpointing & agent state, RAG and vector memory, structured output / constrained tool calling, OpenAI & Anthropic APIs, local models with Ollama
• Frameworks/Libraries: FastAPI, React, Next.js, Spring, Django, Flask, Node/Express, PyTest, NumPy, Pandas
• Tools/Platforms: Docker, Kubernetes, Kafka, Pub/Sub, BigQuery, Terraform, Jenkins, AWS, Google Cloud
• Databases: PostgreSQL, Redis, SQLite, MySQL, MongoDB, Firestore
• Version Control: Git, GitLab, Bitbucket

• PyAutonomy — a local-first autonomous AI agent that runs as a daemon on your own machine: an agentic loop driving 38 tools behind approval policies, multi-agent orchestration, vector-search memory, scheduling, and self-hosted observability. Works with cloud models or fully offline with local ones.
• GATE — an experiment in grammar-constrained tool calling: advertise tools as a compact menu and constrain the arguments, instead of shipping every JSON Schema in the prompt. Big token savings on local models, and surprisingly, better tool selection too.
• Writing — I think out loud about agent engineering and LLM economics on my blog.

• Notion AI: Notion clone with AI built in — DALLE-3 thumbnails and GPT suggestions while you write.
• Langchain_Personal_Assistant: RAG assistant that answers from your own documents.
• BookStore-Protobuf-gRPC: A bookstore service implementing cross-language APIs with Protobuf and gRPC.
• Data-Structures-Python / Algorithms: The fundamentals, written down properly.

• 🔭 I'm currently deep in multi-agent orchestration and the practical side of making agents dependable.
• 🌱 I'm currently learning about grammar-constrained decoding, agent memory architectures, and what long context windows change about agent design.
• 👯 I'm always up for collaborating — agent infrastructure, local-first AI, or honestly any interesting problem where we'd both learn something.
• 💬 Ask me about anything — AI agents, tool calling, full-stack, cloud, or how to get started with any of it. Happy to help.
• 📫 How to reach me: Just follow the links on the top of this page, you know where to find me :D
• ⚡ Fun fact: You will find me playing Valorant or Rocket League while I am not coding.
• 👻 I am known to be a person who gets the work done, and I am always looking for ways to improve my skills and learn new technologies.

Feel free to reach out — for collaborations, questions, or just to say hi!