Work
Recent independent projects across applied AI, geospatial ML, and full-stack platforms. Full case studies — design decisions, stack, tradeoffs, screenshots — live on the portfolio site.
AI-assisted editor that turns a 50,000-asset library of 4K+ video and hi-res photos on self-hosted Immich into a working publishing pipeline.
Color, trim, grade, generate metadata with a local LLM, publish directly to outandaboutwithjim.com, and reconcile the live site's state back into the editor — built around a codec-aware proxy pipeline that survives mixed HEVC, ProRes, and RAW.
Geospatial ML platform for invasive-species monitoring on AWS — from drone orthomosaic to validated training set, with full provenance every step.
Versioned orthomosaics, detection provenance, canonical site identity, lifecycle tiering across S3 storage classes, and reproducible exports that feed the next training cycle. Authenticated production system with field-biologist users.
Personal photo & video gallery — 140+ pieces across drone aerials, landscapes, and travel, indexed on an interactive map.
The publish target for DAM Core. A media-first Django app with HLS video, MapLibre map view, YouTube integration, and a search/discovery UX designed for visual browsing rather than feed scrolling.
Browser-streamed terrestrial LiDAR capture — a colorized Cloud-Optimized Point Cloud, served from S3 + CloudFront with HTTP range requests and rendered with Potree.
Capture preview, hosted on the same AWS account as the rest of the site. The COPC viewer streams only the octree nodes the camera can see, so the wire never carries the whole capture. The v2 of this page lands this summer: a re-scan paired with a small AI change-detection workflow that flags what's different between visits.
The full portfolio — case studies, technical writeups, and live links — is at cessnajim.github.io.