Case studies
Production AI, end to end
Selected work spanning edge agentic AI, multimodal retrieval, inference optimization, LLM evaluation, distributed systems architecture, product ownership, and applied computer vision. Each one is framed around the problem, what I built, and what changed as a result.
Edge AI + Multimodal Retrieval + Agentic Systems
Real-Time Video Intelligence on Edge Devices
Built an on-device system that watches live video in real time and also lets people search and ask questions across everything it has already seen, with both halves running on embedded hardware instead of in the cloud.
- 45% reduction in manual monitoring effort
- 30 FPS sustained real-time analytics on embedded hardware
- Sub-second search across weeks of recorded video
Agentic AI + Enterprise Workflow Automation
Agentic Planning Support for Product and Engineering Teams
Built an agentic workflow that surfaces the dependencies, trade-offs, and resourcing conflicts that normally only show up once a sprint is already underway.
- Dependencies and constraints made visible before commitment rather than after
- Trade-offs surfaced as explicit options instead of implicit assumptions
- Planning context captured once and reused each cycle
ML Systems + Inference Optimization
High-Performance Inference and Deployment for Real-Time AI
Rebuilt the inference and deployment path for real-time AI workloads so latency became predictable, deployments became repeatable, and regressions got caught before they reached a device.
- 2x inference throughput improvement
- 60% latency reduction
- 50% reduction in inter-process latency
Computer Vision + Applied ML
Real-Time Defect Detection for Infrastructure Inspection
Built an inspection system that classifies structural defects from live survey video, so findings arrive while the crew is still on site instead of weeks later in a report.
- 90% classification accuracy
- 4x faster processing
- 30% lower inspection costs
LLM Automation + Developer Productivity
LLM-Based Pull Request Review Assistant
Built a review assistant that reads pull requests and leaves the comments a careful reviewer would leave, so the mechanical problems get caught before a person spends attention on them.
- Mechanical review findings surfaced before human review begins
- Review feedback made consistent regardless of reviewer load
- Human attention redirected toward design judgment
Edge Systems + Distributed Reliability + Python
Edge to Cloud Sync for Unattended Field Devices
Designed and built the service that moves data off unattended field devices and into the cloud reliably, over networks that drop constantly, without ever losing a file or uploading one twice.
- Zero data loss across crashes, power loss, and network drops
- No duplicate uploads under concurrent processing
- No long-lived cloud credentials stored on any device
Full-Stack + Zero-Trust Security + Platform
Multi-Site Visitor Management Platform
Architected and built a visitor management platform for a multi-site organization, replacing manual sign-in with a governed approval workflow, fast unattended kiosks, and a complete audit record.
- Sub-second kiosk check-in and checkout
- Manual sign-in replaced with a governed approval workflow
- Complete auditable record of visits and data access
Distributed Systems + Platform Architecture
One Coordination Decision Across Three Production Systems
Three production systems could each only run one worker, for what looked like three unrelated reasons. They turned out to be the same problem, so I solved it once and applied it three times, then deliberately left a fourth system alone.
- 3 production systems unpinned from a single worker
- 100% in-flight work survival across restarts, up from none
- 5 production systems audited and classified
LLM Systems + Evaluation & Cost Optimization
Making Two LLM Systems Measurable, Then Cheaper
Two production LLM systems could report exactly what a run cost and nothing at all about whether the output was any good. I built the missing measurement first, then used it to make the cost reductions safe.
- 2 production LLM systems moved from cost-only to quality observability
- Conformance checks gating every prompt and routing change
- Ranking weights replaced with measured values instead of guesses
Product Management + Zero-to-One Delivery
BeltIQ
Owned BeltIQ as product manager, from an observation on a site walk to a product running in production, which meant turning an open-ended safety problem into a sequenced roadmap, a hardware supply chain, and a story a buyer could actually evaluate.
- Zero to one, from a site observation to a product running in production
- One roadmap shared across ML, software, hardware, and field teams
- Market-entry strategy and pitch narrative defined before the first client demo