Applied AI/ML research across computer vision, large language model systems, and data intelligence — with a focus on public safety, transportation infrastructure, and education.
Rate intelligence platform for the US trucking industry
Building an AI-powered rate intelligence platform to eliminate the $0.41/mile information gap between freight brokers and trucking carriers. Features an AI floor rate engine that analyzes real-time lane pricing from carrier transaction data to flag below-market offers and generate counter-offer recommendations. Includes broker scorecards ranking 31+ brokerages on pay speed, fairness, and reliability using 28,000+ data points.
$0.41
per-mile gap targeted
Why it matters
Owner-operators and small fleets lose thousands annually accepting below-market rates because they lack pricing data that brokers have. Flowtrate gives carriers the same intelligence edge — turning manual guesswork into data-driven negotiation. This directly addresses a structural power imbalance in a $900 billion industry.
- AI floor rate engine with real-time lane pricing analysis
- Broker scorecards ranking pay speed, fairness, and reliability
- Counter-offer recommendations for below-market loads
- Automated rate benchmarking replacing manual guesswork
Next.jsTypeScriptPythonPostgreSQLPrismaAI/ML
Research in real-time multi-model computer vision for highway incident detection
Investigates the feasibility of automated crash detection across live DOT highway camera feeds using a chained computer vision pipeline — YOLOv8 for vehicle detection, Farneback optical flow for motion analysis, and a custom-trained CNN classifier for crash confirmation. Research covers 50+ cameras across 10 US states, with incident-to-report latency under 30 seconds.
<30s
incident-to-report time
94.2%
crash detection accuracy
Why it matters
This research addresses a documented gap in public safety infrastructure: existing highway cameras generate no automated incident alerts. The multi-model pipeline is a novel approach to real-time detection at scale, with direct applicability to smart city infrastructure and emergency response systems. The work demonstrates that sub-30-second automated detection is achievable with current open-source models and commodity compute.
- Sub-30-second incident-to-report pipeline
- Custom CNN crash classifier trained on real highway footage
- Optical flow analysis for multi-vehicle motion tracking
- PostGIS spatial queries for geographic incident mapping
PythonYOLOv8OpenCVFastAPINext.jsPostgreSQLPostGIS
Research in domain-specific LLM integration for industrial diagnostic systems
Investigates how large language models can be applied to highly specialized technical domains — specifically diesel engine diagnostics in commercial vehicle repair. Research explores retrieval-augmented generation with domain-specific knowledge bases, inventory-aware prompt engineering, and context injection strategies using Llama 3.3 via Groq. Findings are deployed in a production multi-tenant platform.
40%
faster diagnostics with AI
2
platforms (web + mobile)
Why it matters
The research demonstrates that domain-specific RAG with structured context injection outperforms general-purpose LLM queries for industrial diagnostics. This has direct implications for how AI assistants should be designed for expert technical domains — manufacturing, aviation maintenance, medical equipment — where generic model knowledge is insufficient and hallucination has real operational cost.
- 100+ REST API endpoints with role-based access control
- AI assistant for diagnostics and parts recommendations
- Stripe-integrated invoicing and payment processing
- Mobile app for technicians in the field
Next.jsTypeScriptPostgreSQLPrismaStripeGroqLlama 3.3React Native
FMCSA & USDOT compliance intelligence — identifying unsafe and fraudulent carriers before they cause harm
A federal transportation safety compliance platform built to protect the public by identifying bad actors in the U.S. motor carrier industry. FleetSight indexes the full 4.4 million record FMCSA national registry and applies a graph-based entity resolution algorithm to detect "chameleon carriers" — unsafe operators who dissolve their companies and re-register under new identities to erase crash histories, out-of-service orders, and BASIC safety violations. The platform cross-references USDOT numbers, EINs, phone numbers, physical addresses, and named principals to surface fraudulent rebranding patterns that put American lives at risk on public highways.
4.4M
FMCSA carriers indexed
~5,000
annual US crash fatalities
12+
USDOT data points per carrier
Why it matters
Commercial vehicle crashes kill ~5,000 people and injure 160,000+ in the US every year — and FMCSA has no automated system to catch carriers that reset their safety records through identity fraud. FleetSight fills that gap. It gives freight brokers, insurers, compliance teams, and safety regulators the intelligence to identify high-risk carriers at the point of contracting — before a preventable fatality, not after. This is applied AI in direct service of U.S. public safety and federal regulatory enforcement.
- Full FMCSA national registry — 4.4M carriers searchable in real time
- Chameleon carrier detection algorithm with no existing federal equivalent
- BASIC safety score analysis, crash history, and out-of-service records
- Built on public FMCSA and USDOT data in service of U.S. transportation safety
Next.js 14TypeScriptPrismaPostgreSQLTailwind CSS
Research in LLM-driven behavioral analytics for early academic disengagement detection
3rd Place — Education Track · IBM SkillsBuild AI Hackathon · Ohio State University
Conducted with Mitchell Hooper and Jaeha Lee at the IBM SkillsBuild AI Hackathon at OSU, sponsored by IBM, Buckeye FinTech, and BeMyApp. Investigates how passive digital behavioral signals can be analyzed using IBM WatsonX Granite to detect early indicators of student burnout and disengagement — before they manifest as dropout. Research examines FERPA-compliant data architectures for behavioral analytics in higher education.
23
dashboard insight cards
2
user roles (student + educator)
Why it matters
Applies large language models to a critical gap in higher education — early detection of student mental health decline before it leads to dropout or crisis. The privacy-first architecture proves that AI-driven behavioral analysis can be deployed in FERPA-regulated environments. Demonstrates how LLMs can be used not just for content generation, but for proactive human wellbeing intervention at scale.
- 15 student-facing dashboard insight cards
- 8 educator-facing analytics cards
- FERPA-compliant data handling architecture
- Chrome extension for passive behavioral signal collection
Chrome ExtensionReactIBM WatsonXGranite LLMNode.js