Looking for a job

Sri Harsha Mudumba

@sriharshamudumba

Message

I build reliable LLM agents, RAG systems, and distributed ML pipelines.

United States

Message

What I'm looking for

I'm looking to build trusted production AI systems where I can improve LLM reliability, latency, evaluation, and security while working on scalable agentic workflows.

At Procal Technologies, I build Python-based LLM agent workflows, context engineering, and RAG pipelines that connect internal knowledge bases with APIs and tools. My work improved retrieval accuracy by 28%, reduced query latency by 15%, and supported multi-agent communication patterns with 99.8% uptime.

I design evaluation and observability tooling that catches regressions before release, and I optimize Kubernetes inference pipelines for production reliability. I’ve reduced p95 latency by 35%, maintained 99.9% availability under peak loads, and automated CI/CD practices that cut deployment time by 40%.

My foundation includes reinforcement learning research at Iowa State and production data engineering at Cognizant. I’ve trained distributed PyTorch models on multi-node A100 clusters, optimized Python and Oracle data pipelines, and built compliance-conscious systems for PII and regulated insurance data.

Experience

Work history, roles, and key accomplishments

Current

ML Engineer

Current

Procal Technologies

Oct 2025 - Present (11 months)

Owned end-to-end development of LLM-based agentic workflows using Python, delivering scalable orchestration that supported 4 multi-agent communication patterns with 99.8% uptime. Engineered context engineering and RAG pipelines, improving retrieval accuracy by 28% and reducing query latency by 15%.

Python LLM Agentic Workflows Context Engineering RAG API Integration Integration Tools Kubernetes CI CD Test Writing Security Compliance PII Handling Evaluation Tools Error Analysis Statistical Measurement Prompt Architecture AI Coding Agent

MS Thesis Researcher

Iowa State University

Jan 2024 - Dec 2025 (1 year 11 months)

Designed a reinforcement learning agent in PyTorch that balanced accuracy and energy consumption, achieving near 99% prediction confidence. Built a distributed training pipeline using PyTorch Distributed Data Parallel across multi-node A100 clusters, accelerating training throughput by 3x.

PyTorch Reinforcement Learning Distributed Training DDP A100 Reward Shaping GAE Entropy Regularization Gradient Clipping Learning Rate Scheduling Error Analysis

Software Engineer

Cognizant Technology Solutions

Aug 2020 - Jul 2023 (2 years 11 months)

Profiled production Python feature pipelines using cProfile and line_profiler to identify latency bottlenecks, improving pipeline execution time by 25%. Optimized feature engineering pipelines with NumPy and data pipeline design, reducing end-to-end latency by 30%.