I work in NLP, LLMs, and transformers, turning unstructured real-world text into structure through entity extraction, retrieval-augmented generation, and document understanding.
I'm an ML engineer who likes the unglamorous part of NLP: the messy, inconsistent, real-world text that most models choke on. My work blends production engineering with research-driven modeling, from hierarchical transformers to LLM-assisted dataset generation and large-scale text processing.
Right now I'm designing a context-aware transformer that fuses sentence, section, and document-level reasoning for robust employer and entity extraction. I care about systems that ship, reproduce, and hold up on data they have never seen.
A local retrieval agent over PDFs using LangChain, LlamaIndex, and Ollama for question answering and contextual summarization. Runs fully offline.
A retrieval-augmented chatbot that answers questions over uploaded PDFs, built on AWS Bedrock foundation models and served through a Streamlit interface.
A Streamlit app that parses documents and pulls out structured fields using large language models, built for fast and accurate data extraction.
A skincare recommender combining item-based collaborative filtering and content-based filtering, served through a Flask API and an interactive Streamlit app.
An optical-flow plus LSTM system that detects heavy-object anomalies in real-time waste-sorting conveyor footage, improving waste-to-energy processing.
XGBoost on 590K Vesta transactions, reaching ROC-AUC 0.89 and F1 0.76, tuned for the precision-recall tradeoff so real fraud is caught without flooding false positives.
I'm open to ML and NLP engineering roles and to collaborating on applied LLM systems and transformer research.