AI / ML Engineer
Building production LLM systems and measurable ML
Two years shipping LLM and RAG features in enterprise settings. Currently pursuing an MS in Engineering Data Science and AI at the University of Houston.
About
I build production AI systems—LLM applications, RAG pipelines, and evaluation-minded ML—with experience at Accenture and research at Samsung PRISM. My interests sit at the intersection of generative AI, explainable AI, computer vision, and reinforcement learning.
Stack
01
Deep learning, LLMs, RAG systems, explainable AI, computer vision, and reinforcement learning
02
Full-stack development focused on AI application deployment and REST APIs
03
Azure cloud services, ML pipeline automation, and model deployment
Selected work
LaTeX-native, evidence-grounded resume tailoring with FastAPI, Next.js, multi-model routing, anti-fabrication checks, and a reproducible benchmark harness.
Enterprise LLM reverse-engineering platform with function calling and prompt engineering, cutting manual cross-language code migration effort by 70%.
Fine-tuned DistilBERT for 8-class complaint routing, plus a faithfulness audit showing IG, SHAP, and attention rollout agree on only 13–29% of top-attributed tokens.
Comparative study of transformer and graph embeddings for extractive news summarization, evaluated with ROUGE on the BBC corpus.
Deep ensemble of VGG16, MobileNet, InceptionV3, and Xception with data balancing and augmentation for corn leaf disease classification.
Q-Learning, Dueling Double DQN with prioritized replay, and PPO agents navigating a grid world with static and dynamic obstacles, benchmarked against Dijkstra.
Deep Q-Learning agent in PyTorch that learns 2048 from scratch; DQN and Dueling DQN tuned with Optuna, both beating a published baseline.
Research
Deep-learning model that converts voice commands into spectrograms for real-time hands-free maze navigation.
Computer-vision gesture interface for virtual mouse and keyboard control with real-time hand tracking via OpenCV and MediaPipe.
Career
Built scalable AI features for the GenLite reverse-engineering platform using LLMs, function calling, and prompt engineering. Designed a code conversion module that reduced manual migration effort by over 70%. Delivered RAG and chatbot POCs, several of which moved into production.
Researched model interpretability with a team of six using SHAP, CEM, and LIME, improving explainability by 40%. Developed an XAI framework for production ML systems.
Academics
CGPA: 4.0 / 4.0
CGPA: 8.43 / 10
Next
Open to roles and collaborations on production AI and data-driven products. Happy to talk about LLMs, evaluation, or anything in between.
Enjoyed this portfolio?