Freelancer

Summit Interconnect (2024-05-01 – 2025-10-31):

Led Project Echo, a project to utilize LLMs for ITAR classification and information extraction from precision circuit board diagrams as part of Summit’s Alpine Software platform to streamline PCB production

Engaged in extensive prompt engineering with GPT, Llama 3, and Mistral, achieving 99% accuracy, 0.986 precision, and 0.998 recall

Developed and maintained an end-to-end pipeline on Microsoft Azure GovCloud using OCR software and frameworks such as ABBYY and Docling, pre-/post-processing scripts, and API endpoints for PCB diagram classification

Integrated Project Echo into Alpine, resulting in a 50% reduction in latency by replacing manual workflows with LLM-based automation

Built a contextual chat platform with RAG architecture on Azure GovCloud using Azure Cosmos DB and Azure AI Search to provide chat summaries and extract order information

Experimented with GPT, Mistral, and Claude on Azure with datasets managed in pandas, indexed with FAISS, and hosted on Azure Cosmos

Developed a chat summarizer providing summaries in 40 seconds for one year of chat history vs. 120 seconds with Slack’s AI summarizer

VM Consultants (2024-03-01 – 2025-05-31):

Led a team to develop a smart assistant for EOD technicians using a Llama 3-based model with RAG system

Trained RAG models on a 40 GB text dataset for multi-step dialogue, covering data collection, pre-training, and fine-tuning

Implemented dataset pre-processing (OCR/text-scanning, text cleaning, chunking, embedding, indexing with FAISS)

Established a collaboration environment with Atlassian, Slack, and GitHub for team development

Built Raspberry Pi prototypes tested by U.S. Army EOD technicians in field conditions

DataAnnotation (2024-01-01 – 2025-01-31; 2025-10-01 – Present):

Engaged in model training to analyze and evaluate LLM responses for truthfulness and helpfulness

Wrote and corrected model responses to improve accuracy and quality