Onboarding Checklist
What Ag Express brings to the build
The corpus and rules that make the Ag Express platform distinct from a generic diagnostic tool come entirely from your side. This page lists exactly what we need from Eric and the team - organized by the three delivery buckets - and calls out the open questions that need resolution before architecture is finalized.
01
How the three delivery buckets work
Every Brain engagement works the same way. You bring three categories of material. Each one becomes a different class of asset with a different ownership model. The class determines who can export it, modify it, and what happens to it if the engagement ends.
Bucket 1
Domain Corpus
Eric and the Ag Express team bring documents, manuals, and institutional knowledge - everything that makes your diagnostic capability unique.
Class A or B
ICW and Vanguard material may qualify as Class A (platform-owned module). Third-party OEM docs become Class B (your tenant configuration).
Bucket 2
Business Rules + Brand Voice
Defined in a workshop with your team - how verdicts are structured, escalation thresholds, technician communication tone.
Class B tenant config
Prompts, routing rules, and ontology definitions. Owned by Ag Express. Exportable at any time.
Bucket 3
User Data
Live bench session data generated by your technicians at runtime. Never ours.
Class C
Yours always. Deleted on unhook. Technician-approved outcomes can be promoted to corpus - that promotion is your decision, not ours.
02
What Ag Express brings specifically
Each bucket has a concrete checklist. The domain corpus is the most important and the most time-consuming to prepare - it is what makes the Ag Express Brain different from every other diagnostic tool. The business rules can be gathered in a two-hour workshop. User data generates itself once the bench tool is live.
Corpus - what to bring
ICW harness specs + Vanguard product data
First-party manufacturing specs where Ag Express IS the OEM. Highest-value corpus entries - no third-party permission required.
Class A candidate
Equipment manuals
All OEM documentation for brands serviced. PDF, digital, or printed - any format that can be scanned and ingested.
Fault codes + diagnostic trees
Structured or unstructured - existing guides, laminated bench cards, handwritten notes. Everything gets structured by the ingestion pipeline.
Historical repair tickets
The closed-loop training signal. Even partial records are valuable - diagnosis + outcome is what the corpus learns from.
Multi-brand failure taxonomy
Any existing categorization of failure modes by brand, equipment type, or symptom. Senior technician knowledge counts here.
Rules + voice - workshop topics
Verdict structure definition
What does a complete bench verdict look like? Confidence threshold, citation requirement, recommended fix format.
Escalation rules
When should the bench tool hand off to a human technician? What signals an out-of-distribution case?
Repair workflow steps
The standard flow from device intake through diagnosis through quote generation. How it actually runs at the bench today.
Technical communication tone
How should the system speak to technicians? Concise and procedural? Explanatory? This is set in the brand voice config.
User data - generated at runtime
Live bench session data
CAN traces, technician inputs, diagnostic steps taken - generated at runtime. Class C: yours, never ours.
Repair outcomes + technician corrections
When a technician corrects a verdict, that correction is Class C. You can choose to promote approved outcomes to corpus - that's your call.
Quote and work order data
Generated by the bench tool at runtime. Stays on your infrastructure, never leaves your hardware.
03
Open items before architecture is finalized
ICW and Vanguard data rights - Class A scope decision
For ICW-manufactured harnesses and Vanguard products, Ag Express is the OEM. This means the manufacturing specs and failure data have no third-party IP restriction - they may qualify as Class A material (platform-owned module built from your data). This is the highest-value designation and the one that makes the trace library defensible. The scope of what qualifies as Class A needs to be agreed in Phase 0 before corpus architecture is finalized.
Corpus volume and format survey
This is primarily a document ingestion problem, not a model problem. The Phase 0 discovery sprint needs to map every document type, estimated page count, and format before corpus architecture is sized. A 10,000-page OEM library and a 500-ticket repair history are very different ingestion problems. Eric should inventory available material before Phase 0 begins.
Third-party OEM manual licensing
OEM manuals for brands Ag Express does not manufacture (Deere, CNH, AGCO) may carry use restrictions. The July 2026 FTC/Deere settlement expands independent repairer access, but the specifics of what can be ingested into a proprietary corpus need to be confirmed. This affects which documents become Class B versus what requires a different treatment. Can be resolved during Phase 0 corpus mapping.
04
How we start
Phase 0 is a fixed-fee, 2-4 week pre-build sprint. Hardware is identified, procured, and installed before a single line of product code is written. The corpus inventory and Class A scope decision happen here - so the build that follows has a foundation, not a guess.
01
Sign + NDA
Engagement signed. Site access confirmed. Corpus inventory begins immediately.
02
Hardware spec + install
GPU inference server, NAS, and networking scoped and procured in Ag Express's name at cost - no markup.
03
Corpus inventory + rules workshop
Map every document type and volume. Run the two-hour business rules workshop. ICW Class A scope confirmed.
04
Phase 1 kickoff
Hardware live. Corpus pipeline designed. Knowledge graph schema confirmed. Build begins.