AI/ML engineer building agent infrastructure

I build AI systems for messy, real-world workflows.

LangGraph workflows, retrieval and evaluation pipelines, browser agents, automation runtimes, and internal copilot systems designed to be observable, usable, and hard to misread.

01 / Live Systems

Selected agent systems

Built around explicit loops: plan, retrieve, act, observe, evaluate, and recover.

ClosedClaw artifact board with a capsule file tree, run terminal, approval gate, and evidence panel.

framework

ClosedClaw

Local capsule-based agent orchestration with LangGraph routing, sqlite-vec search, approval gates, and JSONL audit trails.

LangGraph sqlite-vec approval gates JSONL audit
ForrestRun artifact board with workflow YAML, runtime terminal output, runs database state, and debug commands.

runtime

ForrestRun

Zero-dependency workflow runtime for YAML-defined agent, function, and tool steps with SQLite checkpoints and replay.

Python YAML SQLite replay
MyCompAgent artifact board with DOM snapshot, Gemini function call, actions log, and Playwright surface.

browser agent

MyCompAgent

DOM-driven browser agent using Gemini native function calls, approval modes, Playwright execution, and run evidence logs.

Playwright Gemini memory debug logs
Aqua Twin artifact board with Streamlit chat, parameters JSON, LangGraph route, and stable-range gauges.

digital twin

Aqua Twin

LangGraph water-management assistant that classifies prompts, extracts simulation parameters, and analyzes stable ranges.

simulation Streamlit LangGraph stable ranges

02 / Experience

Applied in real teams

Internships and production-adjacent work that shaped how I think about agents, retrieval, and evaluation.

03 / Writing

Notes from the build

Longer explanations for agent systems, structured content, and interface decisions.

04 / Background

Signals behind the work

Enough context for recruiters without letting credentials overpower shipped systems.

Education

Vellore Institute of Technology

B.Tech in Computer Science Engineering, 2022-2026. CGPA 9.42/10.0.

Certifications

Applied AI foundations

DeepLearning.AI supervised ML, Hugging Face agents and LLM courses, Google Cloud LLM/GenAI/Transformers.

Leadership

Toastmasters International

President, Vice President of Education, Program Quality Director, and contest wins across speaking formats.

05 / Contact

Work links and contact

Email is best for direct conversations. GitHub and writing are the clearest places to inspect the work.

Email yash.agr1510@gmail.com