Charlie Bushrow

Data Scientist

Business Intelligence Agent: GenAI SQL Assistant

• Built multi-agent workflows combining LLMs, vector databases, and knowledge graphs to interpret domain-specific business context through MCP servers and tool definitions.
• Led research and development of a generative AI assistant that turns natural language questions into executable SQL for non-technical stakeholders.
• Cut project manager dependency on data engineering teams by 20%, saving 120 FTE-hours per week and reducing data request turnaround from hours to minutes.

MyVoice: Enterprise NLP and Sentiment Pipeline

• Owned scientific and architectural direction for an end-to-end NLP pipeline processing 100K+ unstructured comments monthly as the sole supporting data scientist.
• Shipped generative summarization on Claude via Bedrock, using prompt engineering and MapReduce-style chunking to work around token limits.
• Fine-tuned domain-specific models for sentiment detection, multi-class categorization, and PII redaction.
• Hardened the serverless pipeline with schema validation and dead-letter queues, reaching 99.99% reliability on sensitive HR data.

Causal Productivity Modeling and Process Automation

• Built a causal inference model using two-stage least squares regression to measure how operational changes affected employee productivity month over month.
• Productionized a requisition time-to-fill regression model with dynamic interview ratios, cutting median time-to-fill 23% and raising hiring manager CSAT 10%.

Engineering Practice

• Instituted required RFCs for architectural changes and added automated testing to the CI/CD pipeline, reducing deployment rollback rates by 40%.

Lead Data Scientist / Associate

Air Force Readiness Analysis and Decision Support

• Informed a $500 million reallocation of Air Force funding through predictive models and sensitivity analysis, then presented the reasoning to senior military leadership.
• Ran monthly aggregation and analysis of global flight data representing $8 billion in annual defense resource allocation.
• Delivered within GovCloud and air-gapped environments under federal information assurance requirements.

Aerospace Computer Vision and Simulation Modeling

• Raised aircraft classification accuracy from 73% to over 98% across 20+ aircraft types by training on varied perspectives, skews, and lighting conditions.
• Shipped the full application: the Python inference engine and a React front end for uploading images and working with the models.
• Developed agent-based simulations of flight line operations in Python, Java, and AnyLogic so leadership could run Monte Carlo what-if scenarios.

Data Science Instruction and Mentorship

• Served as lead instructor for Intelligence Community engineering bootcamps, teaching applied machine learning and data engineering to 200+ analysts.
• Provided subject matter expertise for data science courses delivered to 1,500+ federal employees as part of the agency's modernization effort.