Girwan Dhakal
Accelerated Master’s student at The University of Alabama building agentic AI in production — retrieval agents, LLM data pipelines, and machine learning research that reaches peer review.
About Me
I am a Computer Science researcher and engineer specializing in Agentic Engineering, Artificial Intelligence, and Data Science. I build LLM systems that survive contact with production — retrieval agents, OCR data pipelines, and the evaluation tooling that keeps them honest — and I publish research on the modeling side. Across internships at Shipt and Alabama Credit Union and two years of machine learning research, that work has halved engineer onboarding time, surfaced five figures in annual savings, and reached peer review.
Projects
Personal AI Persona & RAG Chatbot
A context-aware AI chatbot built into this portfolio. It answers questions about my experience, projects, and research as my digital persona, grounded in a retrieval layer over my own work. You can to see it in action.
ClearPath: AI Medical Financial Assistant
A full-stack medical financial assistant that reconciles bills against EOBs and combines insurance, claims, and Plaid financial data to generate personalized payment and coverage recommendations.
Southern Company Fleet Analytics Platform
A Python and Streamlit analytics platform for network infrastructure lifecycle management, pairing predictive risk models with a GPT-4o assistant. Won 2nd Place at the UA Innovate Hackathon.
Speech Act Analysis
The computational pipeline behind my first-authored research on childhood language development, modeling 60K+ child utterances drawn from multiple longitudinal speech corpora.
Skills & Toolkit
Work Experience
Data Science Intern — AI Platform
Built a LangGraph RAG agent for AI/ML engineers to search internal documentation and manage infrastructure configurations, cutting onboarding and internal support time by 50%. Developed a Gemini-based OCR pipeline on GCP and Snowflake for previously unprocessed shopper receipts, identifying $10K in projected annual vendor savings while expanding tax-recovery coverage to 99%. Tuned Shipt’s product substitution-ranking model with Optuna, lifting Recall@K by 20%.
Machine Learning Researcher
First-authored a research paper applying machine learning to childhood language development across 60K+ child utterances from multiple longitudinal speech corpora, now submitted for peer review. Fine-tuned Google FLAN-T5 on 90K utterances to reduce word error rate from 17.2 to 14.4, and benchmarked DeiT, Swin Transformer, and skeleton-based models on 7 hours of child interaction video for gesture classification.
Data Analyst Co-op
Automated recurring workflows with Power Automate, eliminating 16 hours of manual reporting every month. Built and deployed 5 Power BI dashboards with business stakeholders and led weekly user training, increasing dashboard adoption by 50%. Optimized Microsoft SQL Server ETL pipelines to cut daily refresh time by 30%, delivering fresher data downstream.
Technical Lead
Leading engineering on a roommate-matching platform for housing providers, and secured $2,500 in non-dilutive funding through RiverPitch 2025 and the 2026 Aldag Entrepreneurship Competition. Built and deployed a FastAPI adaptive questionnaire that uses LLMs to turn free-text answers into structured roommate profiles, piloted with 50+ users.
Work Experience
Data Science Intern — AI Platform
Built a LangGraph RAG agent for AI/ML engineers to search internal documentation and manage infrastructure configurations, cutting onboarding and internal support time by 50%. Developed a Gemini-based OCR pipeline on GCP and Snowflake for previously unprocessed shopper receipts, identifying $10K in projected annual vendor savings while expanding tax-recovery coverage to 99%. Tuned Shipt’s product substitution-ranking model with Optuna, lifting Recall@K by 20%.
Education
Master of Science in Computer Science (Accelerated Master’s Program)
Bachelor of Science in Computer Science
Contact
Need a response right away?