Building intelligent systems at the intersection of AI engineering and full-stack development. M.S. Computer Science @ Arizona State University — specializing in LLMs, RAG pipelines, and scalable architectures.
About Me
I'm Krish Patel, a graduate student at Arizona State University pursuing an M.S. in Computer Science (GPA: 4.0/4.0), graduating May 2027. I hold a B.Tech. in ICT from Pandit Deendayal Energy University (CGPA: 9.24/10, May 2025).
I build production-grade AI systems — fine-tuning large language models, constructing RAG pipelines, designing knowledge graphs, and deploying agentic workflows. On the engineering side, I architect full-stack platforms with React, Flask, Node.js, and cloud infrastructure.
My work bridges cutting-edge AI research with real-world impact. As a Graduate Research Assistant at ASU, I build Legal AI multi-agent systems. I've improved model accuracy by 25%, reduced API latency by 20%, and deployed systems at scale on AWS using Docker and CI/CD pipelines.
Technical Expertise
A dual-threat engineer with deep expertise in AI/ML systems and modern full-stack development.
Work History
Building production AI and full-stack systems at fast-growing tech companies.
Portfolio
From AI agents to full-stack platforms — engineering solutions that push the boundary of what's possible.
Multi-agent chatbot that searches insurance policies online in real-time and tells you whether a medicine is covered by your plan — or recommends alternatives that are covered, and which plan to switch to if needed. Every answer comes with cited evidence from live web searches. Built with LangGraph multi-agent orchestration, Tavily for web search, AWS Bedrock as the LLM backend, deployed serverlessly on AWS Lambda. Voice-enabled via ElevenLabs — ask by voice, get a spoken reply.
Benchmarked Qwen3 8B/14B/32B variants on Tau Bench across HPC clusters, tracking response quality, latency, and cost. Analyzed error patterns and built evaluation dashboards to surface failure modes. Improved responses via RLMs, prompt strategies, and multi-agent architectures.
Full-stack career platform with React, Flask, and Neo4j for path recommendations. Implemented secure JWT authentication and modular APIs delivering real-time skill insights. Modeled skill relationship graphs in Neo4j enabling sub-second traversal across 10K+ nodes. Deployed via Docker and AWS EC2, achieving 92% skill-match accuracy across 1K+ resumes.
Full-stack data analytics tool with a dynamic SQL query engine and caching layer for real-time data visualization. Optimized PostgreSQL schema and implemented Redis caching, reducing load times by 35%. Built with Node.js, React, and PostgreSQL, deployed via Docker.
Human-in-the-loop conversational web automation with LangGraph, Playwright, and FastAPI. Designed a LangGraph state machine with intent detection and hard purchase confirmation gate. Implemented MongoDB preference store with adaptive scoring. Added Redis caching layers, cutting query latency 40% over 1K+ requests.
Benchmarked multiple LLM agents on MedAgentBench, CraftMD, and AgentClinic clinical tasks. Ran distributed benchmark jobs on ASU Sol supercomputer with SLURM, Apptainer, and Gaudi2. Evaluated diagnostic accuracy across 1K+ clinical cases, surfacing key agent failure modes. Presented a CREST Phase 2 research poster on multi-agent medical reasoning results.
Seven-agent LangGraph math tutor diagnosing and correcting common student errors. Distinct agent roles for hinting, grading, and feedback using Kiro IDE and AWS. Error-classification engine across 200+ problem types with 90% accuracy. Won the education track at a hackathon with a full spec-driven dev workflow.
Native in-vehicle Android app in Kotlin and Jetpack Compose for infotainment controls. Integrated native C++ modules via NDK/JNI bridges for low-latency hardware signal processing. Configured Jenkins CI/CD pipelines automating builds and tests across 3 device targets. Optimized UI render performance, achieving smooth 60 fps on constrained embedded hardware.
Autonomous multi-agent pipeline for end-to-end ML experimentation and model tuning. Automated hyperparameter search and model selection, cutting research iteration time by 50%. Designed agent orchestration logic for data prep, training, and evaluation handoffs. Pitched to a robotics industry contact, validating real-world research use cases.
Academic Journey
Achievements
Won the education track at a hackathon for Mistake Museum, a seven-agent LangGraph math tutor with a spec-driven dev workflow and 90% error classification accuracy.
Placed top 5 at Innovation Hacks 2.0 for PrismRx, an AWS Bedrock RAG clinical copilot that cut hallucination by 30% and scaled to 500+ concurrent users serverlessly.
Led as Event Management Head of Encode coding club at PDEU for 12 months, organizing and executing 10+ technical events for the student developer community.
Recognized for scalable backend and database architecture across SIP and Odoo Combat hackathons, demonstrating expertise in high-performance system design under time constraints.
Get In Touch
Whether you're looking for an AI engineer, a full-stack developer, or a passionate problem-solver — I'd love to connect. Currently seeking internships, research collaborations, and full-time opportunities.