I build AI tools that help engineering teams move faster.

Former AMD engineer focused on AI infrastructure, AI knowledge search, backend systems, and performance tooling. Looking for a co-op building AI infra or search tooling for engineering teams.

Seeking co-op and internship roles in AI systems, backend infrastructure, AI knowledge search, and performance engineering.

Flagship project Email me Resume LinkedIn GitHub

2 years building production engineering tools at AMD

20-30% measured efficiency gains in AI workloads

260+ code reviews across engineering repositories

Flagship project

vLLM multi-instance benchmarking infrastructure

Why it matters: this setup tests whether one CPU machine can serve more LLM user requests by splitting traffic across isolated vLLM containers instead of relying on a single model server.

Built with Podman containers, per-instance model loading, CPU pinning, nginx request routing, and a benchmark matrix for latency and throughput scores.

View GitHub tool

User requests

nginx load balancer

Podman vLLM #1isolated container
model loaded

Podman vLLM #2isolated container
model loaded

Podman vLLM #3isolated container
model loaded

Benchmark matrix: latency + throughput scores

Request capacity and regression signal

Selected work

Two projects worth slowing down for.

The homepage now highlights the clearest signals for AI infrastructure and AI search. Deeper AMD projects are still available if a reader wants more proof.

Performance Public commits

Faster AI models on CPUs

Situation: A recommendation model was too slow because its AI framework did not natively use the underlying CPU acceleration library.
Task: Close the hardware-support gap without an existing plugin path to build on.
Action: Integrated AMD's ZenDNN library with TensorFlow and fused repeated model steps so the model could do less work per request.
Result: Delivered about 20-30% efficiency improvement at equivalent accuracy, lowering compute cost and improving response time.

MatMul fusion commit Embedding lookup fusion commit

AI knowledge search Leadership workflow

Engineering status chatbot

Situation: Directors and product managers had limited time to track what engineers were working on and whether projects were blocked.
Task: Give leadership a fast project-health view without manually reading Jira tickets, Confluence pages, and engineering updates.
Action: Built a RAG chatbot over Jira and Confluence with a traffic-light status system: red for urgent input, yellow for blockers, green for on track.
Result: Reduced manual status review time by 80-90% and helped leadership spot high-priority blockers faster.

3 more AMD projects ->

Before that

Earlier engineering experience

AMD Co-op, Jan 2024 - Jun 2024: improved model performance by up to 25% using a smaller, faster number format and helped automate nightly performance testing.
Sypto Backend Intern, Mar 2023 - Jul 2023: reduced cryptocurrency selection time by 4x and improved dashboard response time by 50%.
Silver Touch RPA Intern, Jun 2022 - Jul 2022: automated invoice processing with OCR and machine learning, reaching a 95% automation rate.

Contact

Open to co-op conversations.

I am especially interested in teams building AI knowledge products, AI infrastructure, backend platforms, performance tooling, or systems for applied ML.

jauhari.ra@northeastern.edu Resume LinkedIn GitHub