Inflection Point Advisory
Between what AI says is possible and real token burn.
I help product and engineering teams resolve the AI bets that survived the chatbot stage and now have to ship.
Leading AI at Orange Logic. Twenty years at Trimble before that — Senior Engineering Director, board of Trimble Europe B.V. Still shipping production AI.
Independent product work: DAM Core · NatureNet DataHub · Out & About with Jim
Where to start
Most teams start with one decision
You've already explored AI. The next move isn't whether — it's which decision to settle, which loop to fix, or which workflow to actually ship.
Wedge offer
AI Decision Review — pressure-test a single decision
For the AI decision that survived the chatbot stage but still isn't resolved. A working session plus a written recommendation with reasoning, risks, and kill criteria. Done in a week. Credited against a Sprint within 60 days.
Or if you already know which engagement fits
AI bet to commit to
AI Strategy Sprint
For teams about to commit roadmap and budget to an AI bet. Product thesis, technical feasibility, and a roadmap with explicit build/buy/integrate calls.
4–6 weeks · fixed-price
DetailsLive AI that's plateaued
Production AI Review
For live AI features that have stopped improving. Audit of evals, metrics, and the iteration loop. Output is a written diagnosis and a 30/60/90 plan.
2–3 weeks · fixed-price
DetailsOrg adoption that isn't compounding
Operations AI Engagement
For 50–500-person organizations where individual AI use isn't becoming shared capability. Roadmap plus a working pilot — at least one workflow shipped.
4–6 weeks · fixed-price · pilot included
DetailsSelected Work
The work behind the advice
Recent independent projects across applied AI, geospatial ML, and full-stack platforms. Case studies live on the portfolio site.
DAM Core
2026AI-assisted editor that turns a 50,000-asset Immich library into a working publishing pipeline — color, grade, AI metadata, ship to the live site, reconcile back.
- React
- FastAPI
- FFmpeg
- Ollama
- OpenCV
NatureNet DataHub
2025Geospatial ML platform for invasive-species monitoring on AWS — versioned orthomosaics, detection provenance, lifecycle tiering, reproducible exports for the next training cycle.
- AWS
- RF-DETR
- MapLibre
- GDAL
Interactive viewer
Idaho admin building scan
2024Browser-streamed terrestrial LiDAR capture — colorized scan written as a Cloud-Optimized Point Cloud, served from S3 + CloudFront with HTTP range requests, rendered with Potree. Capture preview; v2 this summer is a re-scan with a small AI change-detection pipeline.
- LiDAR
- PDAL
- COPC
- S3 + CloudFront
- Potree
Perspective
Recent writing
Your scan is actually two scans
Reality capture splits cleanly into render work and model work. They use different tools, target different audiences, tolerate different amounts of AI inference, and answer to different parts of the org. Conflating them is the most expensive mistake in the field today.
ReadWe ran Mask3D on a terrestrial scan it was never trained on
An honest field report on running an aerial-trained 3D instance segmentation model against a ground-based LiDAR capture of a small civic building. What worked, what broke, what it means for clients evaluating off-the-shelf 3D AI.
ReadFrom Personal AI to Organizational AI: A Pragmatic Migration Path
Karpathy showed us the architecture for personal LLM knowledge bases. The harder problem is organizational AI adoption. A pragmatic framework for moving from individual experimentation to compounding organizational capability.
ReadCommon questions
Questions I get
The questions that come up most often before a first call. If yours isn't here, ask in the form below.
-
Which engagement is right for us?
Decision Review if you have a single AI decision that's still stuck after the chatbot stage. Strategy Sprint if you're about to commit roadmap and budget to an AI bet. Production AI Review if your AI is live and not getting better. Operations AI Engagement if you're a 50–500-person org and individual AI use isn't compounding into shared capability. If you're not sure, start with the Decision Review — the fee is credited against a larger engagement if you start one within 60 days. -
How are engagements priced?
All engagements are fixed-price, scoped on the discovery call once I understand the shape of the work. The AI Decision Review is the smallest, single-issue engagement; the Strategy Sprint, Production AI Review, and Operations AI Engagement are larger and priced based on scope. No hourly billing, no scope creep. -
How long does the work usually take?
Decision Reviews are done in a week. Production AI Reviews run two to three weeks. Strategy Sprints and Operations AI Engagements run four to six weeks. I deliberately avoid open-ended retainers — they create the wrong incentives on both sides. -
Do you actually ship working code, or just write strategy?
Both, depending on the engagement. Strategy Sprint and Decision Review are advisory — the deliverable is a memo or a roadmap. Operations AI Engagement includes a working pilot — one or two automations actually shipped with the team that owns them. Production AI Review is diagnostic with a written plan; implementation hand-off goes to your team or to a partner. -
How are you different from a large consulting firm?
Two things. First, the work is done by me — not by a partner who sells it and an analyst who delivers it. Second, I still ship production AI systems every week in my day role at Orange Logic, so the advice is grounded in current practice, not five-year-old experience. -
Do you sign NDAs?
Yes — happy to sign yours, or use a simple mutual NDA if you don't have a standard.
Contact
Tell me what you're trying to commit to
A few sentences on the bet, the decision, or the broken loop. I'll respond personally, usually within a day or two. No funnels, no scheduling tools.