Mangesh Raut

Hi, I’m Mangesh Raut.

Software Engineer · Building Backend, Cloud & Agentic AI Systems

I build software across backend engineering, cloud infrastructure, full-stack products, and increasingly, agentic AI systems. Over 6+ years of hands-on engineering, I’ve worked on production services, databases, infrastructure automation, ML systems, and AI applications.

I care about the parts of engineering that become important after the demo works: reliability, performance, testing, observability, failure handling, and whether the system can actually be measured.

My engineering journey started in Pune, India, where I studied Computer Engineering before completing my B.E. in Computer Engineering from Savitribai Phule Pune University. I later moved to Philadelphia and earned my M.S. in Computer Science from Drexel University in 2023.

My professional experience has taken me through different layers of the stack.

At Harshwardhan Enterprises, I worked in network engineering. At Drexel University, I worked with database systems. During my Database Administrator internship at Aramark, I automated AWS workflows with Python, migrated 3+ legacy databases to AWS RDS, and optimized SQL queries and indexes, improving report execution time by approximately 30%.

At IoasiZ, I worked as a Software Engineer on backend and distributed systems, including Java/Spring-based services, APIs, databases, caching, testing, and production reliability.

More recently, much of my work has moved toward AI systems engineering.

I build systems where LLMs interact with retrieval pipelines, tools, application state, APIs, and real-time interfaces rather than operating as isolated chatbots. My portfolio itself is one of those experiments: an agentic application combining FastAPI, OpenRouter, streaming responses, browser-native WebMCP tools, monitoring, and deterministic local actions.

Building these systems has changed how I think about AI. I’m increasingly interested not only in what a model can generate, but in questions such as: When should an agent reason instead of executing deterministic software? How do we evaluate tool use? What happens when an agent loses state or a tool fails? When does retrieval genuinely improve an answer? How should we trade quality against latency and cost?

Those questions have pushed me deeper into applied AI/ML experimentation. My work includes RAG and retrieval evaluation, speech processing, multimodal applications, agent protocols, model routing, and automated evaluation. I’m particularly interested in the intersection of AI Agents and Systems for AI—building AI systems that are useful because the engineering around the model is dependable.

I also worked on Real-Time Face Emotion Recognition, using Python, OpenCV, and machine learning to build a facial-affect recognition pipeline that achieved 95% held-out test accuracy. That work was an early introduction to the difference between getting a model to run and evaluating whether it actually works.

Outside engineering, I enjoy travel, photography, exploring new technology, and discovering local food. Those interests show up throughout this portfolio as well—it is intentionally more than a collection of project cards. It is a living record of what I build, what I measure, what fails, and what I learn next.

I build things, measure them, break them, and make the next version better.

At a glance

The same facts, compressed for a hiring scan.