I take complex, messy problems and turn them into AI systems that run reliably on their own — from LLM pretraining research to production document-intelligence pipelines.
I'm an AI Engineering Intern at ApolloMD, and I just completed my MS in Robotics and Autonomous Systems (Artificial Intelligence) at Arizona State University.
What drives my work: I don't settle for the first thing that works. On an LLM pretraining research project, an initial approach got us ~15% improvement — I dug deeper, discovered ELECTRA's generator-discriminator method on my own, and pushed the result to 96.42% discriminator accuracy. At ApolloMD, I built a document-processing pipeline that doesn't just extract data — it knows when to trust itself and when to flag something for a human to check, calibrated against real evidence, not guesswork.
I also build AI outside of formal work because I genuinely enjoy it — including Jarvis, a personal voice assistant where I learned hands-on prompt engineering from scratch, and AirBench, a fine-tuning experiment tracker I rebuilt from scratch to genuinely understand production ML infrastructure, load-tested at ~75 req/sec with 0% errors.
🔭 Currently focused on: Forward Deployed Engineering, AI Implementation, and Applied AI/ML roles 🌱 Currently exploring: RAG systems, agentic workflows, and production LLM deployment 💬 Ask me about: LLM pretraining, prompt engineering, robotics decision-making (search/planning/RL), computer vision
Languages & Core
AI / ML
Infrastructure & Web
Robotics & Vision
Tools
|
A voice assistant combining speech recognition, an LLM backend, and system-level automation. Iteratively engineered system prompts for consistent, reliable behavior, plus a local-first routing layer that handles simple requests instantly offline.
|
Designed a multi-task pretraining framework improving logical reasoning in transformer-based LLMs. Discovered and implemented an ELECTRA-style approach on my own initiative, achieving 96.42% discriminator accuracy.
|
|
FastAPI backend for experiment tracking, job orchestration, and real-time metrics streaming. Pluggable multi-provider compute architecture; verified end-to-end with a real subprocess training run and load-tested at ~75 req/sec, 0% errors, 15ms median latency.
|
Modular robotics AI framework integrating search (A*, UCS), PDDL symbolic planning, and Q-learning. Benchmarked: A* cut node expansion 40% vs. BFS at 100% optimal-path accuracy; Q-learning converged 0% → 100% goal-reach over 500 episodes.
|
More projects — vision-language navigation research, a transformer built from scratch, and vision-guided robotic arm control — on my repositories page.
2026 – Present AI Engineering Intern @ ApolloMD 2026 MS Robotics & Autonomous Systems (AI), Arizona State University 2026 Graduate Service Assistant, ASU School of Engineering 2025 Technical Assistant, ASU Thunderbird School of Management 2024 B.Tech Artificial Intelligence, Mahindra University 2023 Software Development Intern, BWS Solutions
I'm actively exploring Forward Deployed Engineering, AI Implementation, and Applied AI/ML roles. Always happy to talk AI, robotics, or anything you're building.
