Hi, I'm Yogesh! I study CS Systems at Stanford and spend most of my time thinking about AI infrastructure and scalable systems.
Find me on GitHub, LinkedIn, or shoot me an email at yogesh60@stanford.edu.
Experience
-
SWE Intern · Meta
I worked on AI infrastructure at Meta, on the event-driven scheduler that allocates server and GPU capacity across a fleet of about a million machines. Added deficit events in C++ so the scheduler notices reservations that stay short of capacity and runs a focused solve for them, shipped a lease-based transfer protocol for moving multi-node GPU reservations that gives the destination first claim on freed capacity, and built a Python/TypeScript debugger that joins scheduler telemetry so on-call engineers can see why a drain or solve is stuck.
-
SDE Intern · Amazon
I built infrastructure for AI agents on Amazon's internal service-to-service identity platform, letting agents call internal services under their own identity instead of impersonating users. Implemented the OAuth 2.0 client-credentials and token-exchange flows, a multi-region client registry on DynamoDB, and an MCP server for programmatic credential management.
-
ML Systems Researcher · Stanford — Prof. Alex Aiken's group
I worked on automated GPU kernel optimization, targeting a hot attention kernel in SGLang on H100s. Built a multi-agent LLM loop in Python that proposed CUDA variants and iterated on Nsight Compute profiles. Productionized the best candidate for a meaningful end-to-end inference speedup. Led to a NeurIPS workshop paper.
-
Enterprise SWE Intern · State Farm
Migrated an on-prem claims service to AWS serverless (Lambda, API Gateway, Step Functions, DynamoDB) in Python, provisioned with Terraform. Standardized GitLab CI/CD across a dozen application teams.
-
Systems Intern · Verizon
Built a REST API comparison tool in Python using DeepDiff to catch response mismatches between staging and production across the team's internal services. Deployed it on EC2 with CloudFormation and set up an internal status dashboard so the team could monitor fleet health without any SSHing.
-
ML Intern · Collier County Public Schools
Built a student-facing chatbot for academic and mental-health support and a companion teacher portal for accessing student records and testing results, across a React / Node / MongoDB stack. Fine-tuned BERT — a transformer-based foundation model, state-of-the-art at the time — for intent classification over district-specific queries, and wired it to per-student context so responses pulled from each student's own records.
-
Coding Instructor & Mentor
Taught machine learning across every audience I could reach in Southwest Florida, middle and high school students, and working software engineers in the area, and wrote and published a small neural net library on PyPI that I used as teaching material.