Basavraj Chinagundi

I like messy backends. I specialize in taking fragile, duct-taped workflows and refactoring them into systems that actually survive production. Whether it's optimizing slow SQL queries or wrangling unpredictable LLM outputs, I focus on building infrastructure that is boring, reliable, and hard to break.

I’m actively seeking Summer 2026 internships in SWE/ML/Backend roles.

Education

M.S. Computer Science - Arizona State University Aug 2025 - May 2027 • GPA: 4.11 / 4.0

Previously: B.E. Electronics & Communication, Thapar University (2019-2023).

Technical Skills

Experience

Deloitte - AI Software Engineer

Oct 2023 - July 2025

I focused on modernization and automation, effectively automating myself out of the boring parts of the job.

• The "Transpiler" Project: Wrote a Python utility that parsed XML mappings and auto-generated optimized SQL logic. It handled a migration of 200+ workflows and killed 90% of the manual grunt work.
• Vector Search at Scale: Architected a FastAPI microservice for semantic search over unstructured docs. Used custom chunking to slash P99 latency by 40%.
• Agentic Workflows: Orchestrated multi-step AI agents using LangGraph, handling state management and conditional branching so the bots didn't get lost in loops.
• Quality Gates: Introduced a strict testing culture with Pytest and Azure DevOps, because I hate waking up for hotfixes.

Samsung SDS - Software Engineer Intern

Mar 2023 - Jun 2023

• Full-Stack Inventory: Built a Spring Boot + React system to manage 1,000+ SKUs, killing a legacy spreadsheet process.
• Performance Engineering: Tuned MySQL schemas and indexes to sustain sub-200ms response times under high concurrency.
• Forecasting: Implemented ARIMA and Prophet models to help the business predict demand rather than guess at it.

Projects

ShadowSearch - 1st Place, Sunhacks 2025

View on GitHub (opens in a new tab) A Chrome extension that turns the current webpage into an interactive knowledge base.

• The Tech: JavaScript, Cloudflare Workers, Llama 3.1.
• The Logic: Ingests active tab content to enable real-time Q&A regarding the page context and generates semantic recommendations for similar reading materials.

Slapp-AI

View on GitHub (opens in a new tab) A multimodal recommendation engine that actually understands what you're looking for.

• The Tech: Python, Vector DBs, Streamlit.
• The Logic: Indexes 5,800+ items and uses vector similarity search to rank results based on evolving user preference graphs.

STaR: Self-Taught Reasoner

View on GitHub (opens in a new tab) Reproducing deep research to understand how LLMs "think."

• The Result: Fine-tuned Llama 3.2-3B on GSM8k, achieving 73.8% accuracy.
• The Goal: Validating the "chain-of-thought" bootstrapping methodology by building the training loop from scratch.

Connect

Email: bchinagu@asu.edu
LinkedIn: linkedin.com/in/basavrajchinagundi (opens in a new tab)
GitHub: github.com/raj-chinagundi (opens in a new tab)