Work
Production systems, supporting builds, and lab notes
Tier 1 is flagship case studies. Tier 2 is strong supporting work. Tier 3 is the archive and lab so the strongest work is never diluted.
Flagship case studies
BearcatGPT
Campus scale AI assistant and agent platform
Production generative AI assistant serving 46,000 plus students and faculty at University of Cincinnati. RAG pipelines on Azure AI Foundry with hybrid retrieval, Socratic tutoring agents, agentic routing, and MCP integrations to external enterprise systems.
TAMALE AI Extraction Gateway
Enterprise RMS integration on a $13 per month footprint
Stateless FastAPI middleware that bridges Advent TAMALE and BearcatGPT. In memory bitstream extraction turns binary research attachments (PDF, DOCX, XLSX, PPTX) into structured JSON envelopes for LLM tool calls, hardened with API key auth, Key Vault secrets, SHA256 hashing, and standards compliant error codes.
AI Meeting Intelligence Platform
Executive use case tracker over Microsoft Teams transcripts
Five layer enterprise workflow that ingests Teams .vtt transcripts, generates strictly grounded JSON via Azure AI Builder, persists to a flat SharePoint list, and exposes the corpus to BearcatGPT through Microsoft Graph. Deliberately skipped a vector database in favor of a per person and date range access pattern.
LectureMind
Full stack generative AI study platform
Generative AI learning platform that turns long form lectures into grounded outlines, summaries, flashcards, quizzes, mind maps, and faculty accessibility reports. Multi model orchestration with fast and reasoning paths, separate student and faculty workspaces, and a transient faculty session mode with cleanup on signout.
Supporting work
Braden GM Morning Scorecard
AI dealership analytics dashboard with privacy by design
72 hour take home for Braden Auto Group. Parses ~5,000 dealership sales rows fully in the browser, computes 14 KPIs plus 5 chart cuts, and sends only aggregated JSON to a serverless Claude route to generate a structured findings panel and a morning briefing in dealership voice.
Make.com Sales Insight Automation
No code AI pipeline from Google Sheets to Claude to email
A no code Make.com scenario that watches new dealership sales rows, sends each row to Claude with a business analyst prompt, and emails a concise insight within seconds. Companion piece to the Braden scorecard.
Secure Sight
Real time weapon detection platform
Best Security Hack winner at MakeUC 2024. Fine tuned YOLOv11 on a 20,000 image dataset, integrated S3 for timestamped evidence, and used Twilio to push WhatsApp alerts with detection metadata to security personnel. Added a web interface for anonymous community incident reporting.
BearcatGPT Operational Agents
Applied prompt engineering at enterprise scale
A grouped case study covering the operational and academic agents beyond the three Socratic tutors: an ACCT3073 Exam Practice Coach with a mastery threshold and jailbreak protection, Huddlemate live interview coaching, an invoice reconciliation agent for Campus Services Inclusive Access, and a utility bills extraction agent.
AI Platform Evaluation and Benchmarking
Vendor scoring and model benchmarking for senior leadership
Structured comparison of Open WebUI and nebulaONE across functionality, security, accessibility, and total cost of ownership, scored and delivered as a leadership report. Includes a multi model benchmarking workbook and a Socratic agent test suite.
Archive and lab
Smaller pieces, side projects, and ongoing exploration. Kept here so the flagship work stays uncluttered.
Invoice and GL Data Processing Agent
Excel in, Excel out reconciliation for Inclusive Access
Merges multi tab invoice and general ledger files, derives course and key fields, applies store mapping rules, and produces summary tabs. Validation checkpoints and row count and sum totals are printed so the model self catches data loss.
Customer Churn Prediction Model
Logistic regression on Telco Customer Churn
Feature engineering, class balancing, outlier handling, and a full evaluation suite including confusion matrix, ROC curve, and feature importance.
Horse Racing Analytics
Ensemble modeling with Kelly Criterion bet sizing
Benter style multinomial logit, weighted power score, and Bolton Chapman value filter with Kelly Criterion bet sizing. Analytical, personal interest work.
MCP Server Exploration
Hands on Model Context Protocol prototyping
Lab work building MCP servers across STDIO and Streamable HTTP transports, using FastMCP patterns to connect AI agents to external data and tools.