AI Engineer

I architected a six-node LangGraph text-to-query pipeline for MCI's SOC team, covering translation, generation, syntax checking, reflection, judging, and execution. The pipeline used immutable Pydantic state and YAML-based configuration for models, prompts, components, and node composition, enabling accuracy and cost comparisons across 30 scenarios without code changes.

I designed dynamic RAG and Self-RAG workflows using pgvector-based retrieval over curated Splunk examples. This improved reference-match accuracy by 23 points over the no-RAG baseline and improved query-generation accuracy by 12% over static few-shot prompting through leave-one-out evaluation.

I implemented bounded query self-correction, live syntax validation, LLM-as-a-judge scoring, execution-based validation, Langfuse tracing, prompt versioning, and checksum-based dataset versioning. I also built a multi-agent enterprise chatbot as the sole developer, enabling natural-language access to PostgreSQL, MongoDB, ChromaDB, MinIO, and parsed enterprise documents.

I engineered prompt optimization strategies using Chain-of-Thought, Tree-of-Thought, Graph-of-Thought, and few-shot prompting to improve reasoning quality and response consistency for medical AI use cases.

I built RAG pipelines that embedded curated medical datasets in ChromaDB for semantic search and context-aware responses. I collaborated with research, engineering, and stakeholder teams to evaluate LLM outputs, refine prompts, and improve task performance.

I designed a dual-branch DistilBERT and temporal CNN architecture with a speaker-metadata embedding network for six-class fake-news classification on the LIAR benchmark. The model achieved 94.5% accuracy and outperformed cited prior approaches while reducing training time.

I built automated ETL pipelines with Flask and REST APIs to collect and process real-time Twitter and Reddit data for NLP model training. I also developed preprocessing workflows and applied sentiment analysis, named entity recognition, and text classification to unstructured datasets.

GPA: 3.89/4.00 (19.44/20). Thesis: Fake News Detection Based on Convolutional Neural Networks (CNN).

GPA: 3.45/4.00 (17.24/20).