CR Tech Lab
Fractional AI architecture · Richmond, VA
I'm Will Cassell, a former data science & analytics leader who now builds AI agents and systems hands-on. I find the business problem first, then build the system that fixes it.
If there's a fit, we'll scope an audit on a 30-minute call. If there isn't, I'll tell you that.
Background on LinkedIn ↗01 · The Gap
I spent my career running analytics teams, so I know how new systems get adopted or shelved. Now I build the systems myself.
The work included budgets, org change, and vendor selection. I build daily and hands-on: LLM agents, voice systems, data pipelines, and workflow automation. When I recommend something, it's because I've already built a version of it.
02 · The Engagement
A fixed-fee, three-week engagement on one department or process. You leave with a ranked roadmap and a working pilot of the top opportunity.
Interviews with your team and a workflow map of the target process: where time, money, and errors accumulate.
Every opportunity scored on ROI against feasibility. Then I build a working pilot of the one at the top of the list.
A 90-day roadmap, the live pilot demo, and a governance and risk snapshot covering data handling, vendor exposure, and AI-use guardrails.
03 · Recent Builds
A plain-language assistant on a company's live job data. A frontline rep asks what's overdue or next and gets their own worklist and checklists. A manager switches roles to see the whole operation, which region's behind and which client's at risk, and can push an update back into the system. The answers are lookups against the data.
Proof Sample field-service operation: 50 staff, 800 jobs, 2,800 tasks · shown live on the call
Hundreds of pages of scanned factory manuals turned into a reference tool: pick a part or describe a symptom, and get the answer, the service steps, and a link to the exact manual page it came from. Every answer is cited to that page. The proof case is a 1970s farm tractor. The same pipeline fits any equipment, maintenance, or SOP library.
Proof 330+ manual pages indexed, 160+ parts, 50 diagnosed symptoms · live at mytractors.app ↗
Applicants who've already applied get a short, structured screening call from an AI that says it's an AI. A few yes-or-no qualifiers (availability, work authorization, physical requirements), with the transcript and a PASS or REVIEW flag waiting for your recruiter. A person makes every hiring decision.
Proof Test calls completed end-to-end over the phone · screening flow shown on the call
Pipelines that pull messy public permit records from multiple cities, match the same contractor across spellings and jurisdictions, and produce a ranked leaderboard of who is active for a sales or underwriting team.
Proof 1.2 million real permit records, two cities · runs offline on the call
I'll demo any of these live on the first call.
Side projects I've built and run myself.
One shelf photo checked against a catalog, with every AI judgment, box and cent of cost visible.
Richmond in 3D, rebuilt from 63,750 building footprints, LiDAR and historic maps.
Virginia's K–12 curriculum broken into connected knowledge blocks you can explore in 3D.
Free doubles pickleball with side-out scoring and a packed stadium, in the browser and on the App Store.
04 · After the Audit
When the audit surfaces more than one thing worth building, I stay on as your fractional AI lead, owning the roadmap, building the pilots, selecting the vendors, and upskilling your team on a monthly retainer.
I take on a small number of retainer clients so the work gets my direct attention. I currently have room for one more.
05 · About
For 25+ years I led data science and analytics organizations, most recently at the executive level: the years where you learn that technology fails for organizational reasons more often than technical ones.
Now I work as an independent architect because I like the craft: finding the business problem, then personally assembling the AI, analytics, and automation blocks that solve it. CR Tech Lab is the practice that work lives under.