Open to full-time AI/ML Engineer roles
I build AI systems that hold up outside the demo.
I'm Thayaa, an AI Engineer in Tempe, Arizona. I build LLM agents, RAG and agentic automation at qBotica, and I'm going deep on GPU programming and LLM inference.
M.S. Robotics & Autonomous Systems (AI), Arizona State University · GPA 3.8
A mock agent console that turns a plain-language request into a validated automation workflow, then sends it to human review.
TOOLS I SHIP WITH
PythonPyTorchHugging FaceLangGraphLangChainFastAPIFAISSRayDockerAWSGoogle CloudC++GitPythonPyTorchHugging FaceLangGraphLangChainFastAPIFAISSRayDockerAWSGoogle CloudC++GitPython, PyTorch, Hugging Face, LangGraph, LangChain, FastAPI, FAISS, Ray, Docker, AWS, Google Cloud, C++, Git
About me
I'm an AI Engineer who builds LLM systems that hold up outside the demo. I finished my M.S. in Robotics & Autonomous Systems (AI) at Arizona State University, and I work as an AI Intern at qBotica in Tempe. There I build the backend for qubi Autopilot, which turns plain-language requests into executable automation workflows.
Before that I built a distributed semantic search and RAG system on ASU's Sol supercomputer for a multi-year research study. Earlier, at RBG.AI, I fine-tuned a quantized 7B LLM with QLoRA on a single T4 GPU for a medical chatbot that went to production.
Now I'm going deeper into the layer under the models: GPU programming, kernels and LLM inference performance. I like problems with real constraints, like limited memory, IP and compliance rules, latency budgets, and a client waiting on a deadline.
0.0
M.S. GPA, Arizona State University
0B
LLM fine-tuned with QLoRA on a single T4
Sol
ASU supercomputer I built distributed RAG on
Real constraints, clean systems.
LLM AgentsRAGFine-tuningEvaluationPython BackendsInference
LLM agents & automation
Agents that turn plain language into reliable, validated actions.
RAG & retrieval
Search over messy, real data, from FAISS on HPC to knowledge graphs.
Fine-tuning & quantization
Adapting open models to a domain under tight GPU memory.
Voice agents
Real-time phone agents with guardrails that hold over long calls.
ML pipelines & evaluation
Data prep, training and evals that are reproducible.
Learning now
GPU & inference
CUDA, HIP and LLM inference performance.
Selected work
Case study · Work
qubi Autopilot (qBotica)
Backend for an agentic automation platform that turns plain-language requests and requirement docs into executable workflows. I replaced an unreliable Computer Use agent with a deterministic pipeline (natural language → JSON IR → schema validation → human review), and designed an IP-safe split where an external LLM drafts a generic plan and a local model maps it to proprietary components.
PythonFastAPILLMsLocal modelsMCPJSON Schema
Case studyCase study · Research
Distributed RAG on a supercomputer (ASU)
Semantic search and RAG over a professor's multi-year, 7-site study on human generosity. Ray parallelises embedding, FAISS handles retrieval, and it all runs on ASU's Sol supercomputer. The research team still uses it.
PythonRayFAISSRAGHPC (Sol)
Case studyTeam project
ATLAS-X: fraud detection with graph features
XGBoost with graph features from a Neo4j card–device–email network, trained on 590k real IEEE-CIS transactions (0.957 AUC, ~17 ms inference). Kafka streaming, FastAPI, Prometheus/Grafana monitoring and LLM-generated explanations.
XGBoostNeo4jKafkaFastAPIGrafana
ExploreEchoShield: on-device voice isolation
Real-time audio system for voice isolation and smart ambient-mode triggers, using WebRTC VAD, Resemblyzer speaker verification and Vosk wake-word detection. Tuned frame size from 30 to 60 ms to fix CPU overload.
PythonWebRTC VADResemblyzerVosk
Explore
+ EXPERIENCE
Jun 2026 – Present · Tempe, AZ
qBotica
AI Intern
- Building the backend for qubi Autopilot, which turns plain-language requests into executable automation workflows through a deterministic, validated pipeline with human review.
- Designed an IP-safe split between an external LLM and a local model; led a 3-person team on a client voice agent with rules-as-code guardrails; built Python backend services for a healthcare data-migration client.
Sep 2025 – May 2026 · Tempe, AZ
Arizona State University
Graduate Researcher
- Built a distributed semantic search and RAG system over a multi-year, 7-site research study using Ray and FAISS on the Sol supercomputer; still in use by the research team.
Dec 2024 – May 2026 · Tempe, AZ
Arizona State University
Text Editor & QA Engineer
- Edited content and performed QA testing, finding bugs and giving feedback.
Nov 2023 – Jun 2024 · Coimbatore, India
RBG.AI
AI Intern
- Fine-tuned a quantized 7B LLM with QLoRA on a single AWS T4 GPU for a medical client's chatbot, deployed to production; built NLP pipelines with Hugging Face and PyTorch.
Aug 2022 – Oct 2022 · Coimbatore, India
Gramophone Media Lab
AR Intern
- Built an AR e-brochure showcasing Vedic heritage with web AR.
+ EDUCATION
2024 – 2026 · GPA 3.8
Arizona State University
M.S. Robotics and Autonomous Systems (Artificial Intelligence)
2020 – 2024 · Coimbatore, India
Sri Ramakrishna Engineering College
B.Tech, Artificial Intelligence and Data Science
+ SKILLS
Languages
PythonC++SQLJavaScript/TypeScript
AI / ML
PyTorchHugging Facescikit-learnXGBoostLangGraphLangChainQLoRA/PEFTFAISSChromaDB
Systems & Data
FastAPIDockerRayKafkaNeo4jStreamlitGitHub Actions
Cloud
AWSGoogle Cloud
+ NOW LEARNING
The layer under the models, where my work is heading.
Learning
CUDA
PMPP book + GPU MODE lectures
Learning
AMD HIP / ROCm
Porting CUDA-style kernels to AMD GPUs
Learning
Modern C++
Memory, RAII and performance for systems code
Learning
LLM inference
KV cache, FlashAttention, batching