AI
Charles Trinh — AI Vendor Due Diligence | Las Vegas, NV · Las Vegas, Nevada · Henderson · Clark County, Nevada · North Las Vegas, Nevada

A vendor's AI features are demonstrated in minutes and evaluated over months. Diligence is what converts the demo into an evidence trail you can rely on before signature and after deployment.
What due diligence is for
Due diligence on an AI vendor is not a checklist of documents. It is the work of finding out what the vendor actually does with your data, what its model does on your task, and how it will change over the term of the contract. The documents are the vendor's claims; the questions and the observed behaviour are your evidence. I treat due diligence as a short, focused evaluation, and I write the criteria before the vendor meeting so the meeting produces evidence rather than enthusiasm.
The first pass is orientation. I read the vendor's product page, their terms, their security page, their subprocessor list and their model documentation. Where the vendor publishes an evaluation report, I read the evaluation and check what it measured. A published evaluation is usually accurate about what it measured and silent about what it did not. That silence is the more informative part, and it tells me what to ask about next.
The second pass is the conversation, and the first question is not “tell us about your model”. It is “describe the data flow for a request from our system to yours, including every subprocessor”. A vendor that can answer that clearly is a vendor that has thought about procurement in healthcare. A vendor that answers with a marketing diagram is unprepared, and the unpreparedness is often a sign of a product-market fit aimed elsewhere.
The data handling evidence pack
Every AI vendor should be able to provide a written evidence pack before the BAA is signed. It should contain the BAA and the subprocessor scope, the data classification by request type, the storage locations for prompts and outputs, the retention period by data class, the deletion procedure and its verification, the encryption details, the access logs and the incident notification procedure. Where any one of these is missing, that is a finding and it is a condition of the contract, not a follow-up after signature.
The BAA is where the scope is decided. A vendor BAA that covers “the software” but not “the model router on which the software depends” is an incomplete scope, and the same applies to “support” and “observability”. I read the BAA against the data flow and I ask for the subprocessors list at the same time. A subprocessor list that changes without notice is a contractual problem that should be a clause, not a policy, and the clause should name how the customer is informed and how quickly.
Some vendors will say they do not require a BAA because they receive no PHI. That is sometimes true and usually not. When it is true, the product should have a documented de-identification step that I can see. When it is not, the BAA is required even for a production inference call that is otherwise stateless. The absence of a BAA in those circumstances is a hard stop for me, and I have ended diligence on that basis more than once.
Model change and version lifecycle
A vendor's AI product is not a fixed product. The model underneath changes on a schedule the vendor controls, sometimes with a pinned model id, sometimes with a rolling update. Diligence should ask for the vendor's model change policy, the notice period, the deprecation timeline and the contract term for a pinned model. Where a vendor has no change policy, the purchase is buying a model of unknown future behaviour at today's price, and the buyer should understand that before signing.
I ask for the vendor's changelog and the last twelve months of it. A model that changed six times in a year without any notice is a model whose current evaluation is worth what it measured in the past. The changelog should distinguish model changes, prompt changes, guardrail changes and interface changes, and it should tell me how the vendor evaluated each. Where the changelog is marketing, I ask for the real one and whether the customer base sees it.
Evaluation and monitoring belong in the diligence. A vendor's evaluation report should cover the intended use, the population and the stratification. Where the vendor has no evaluation programme for a model that influences a clinical or billing decision, that is a material gap. Where it has one, I look at what it publishes — the report itself, the recording of changes, the standing comparison against a frozen reference — not the marketing summary of the report.
Measured evaluation on your task
Documentation and conversation are necessary but not sufficient. The most informative part
of diligence is a measured evaluation against your own data before signature. It is usually
small — a hundred to two hundred samples — and it produces a result that no amount of
vendor claim can replace. I build the evaluation as in
ai-evaluation-benchmarking, with a representative sample and a ground-truth set
from your clinicians. The vendor provides a demo instance, and the evaluation is scored
with the same harness used in production.
Where the vendor will not support an evaluation before signature, that is a finding. The terms can restrict the evaluation sample's use, and I will accept an evaluation with restricted use where the restriction is reasonable. Where the vendor refuses any evaluation at all, the verification is limited to their own claims, and the risk of a mismatch between claim and performance on your task is carried by you. That difference is worth paying for in the negotiation.
The evaluation should test one or two of the vendor's weakest assumptions. If a vendor claims a high accuracy rate on summarisation, I evaluate on long encounters. If a vendor claims support for a particular specialty, I evaluate on that specialty's notes. If a vendor claims BAA-safe processing, I test the BAA against the data flow that the evaluation required. Weak points are the ones that turn into contract changes or production incidents after go-live, and the point of diligence is to find them before signature.
The scorecard and the recommendation
Every diligence produces a scorecard with explicit criteria and explicit scores. The criteria are the evidence pack, the evaluation result, the change policy and the contractual protections. Each is scored against the criteria agreed with the organisation before the meeting, and the scorecard links each score to the document or the measurement that supports it. A scorecard without a document behind it is a summary of impressions, and that is not diligence.
The recommendation follows the scorecard. Where the criteria are met, the recommendation is to sign and the conditions the diligence found become contract clauses. Where the criteria are not met, the recommendation names the gaps and what the vendor would need to do to meet them. Where the gaps are structural — no BAA, no subprocessor list, no change policy — the recommendation is usually not to sign, and I have given that recommendation in several diligences where the demo was excellent.
The scorecard continues past signature. The criteria that were evaluated at signature become criteria the vendor must re-prove on a schedule. Where the vendor's change policy was silent, the contract adds a clause requiring change notice. Where the evaluation was scored on your data, the model change policy should trigger a re-evaluation on the same sample. Diligence is the start of the relationship's evidence, not the end of it.
Procurement in the larger programme
A vendor diligence is one input to the larger AI programme. The procurement decision for a vendor also has budget, workflow fit and staff adoption behind it, and the diligence score is not the only factor. What diligence does is put a floor under the decision: any vendor that fails the floor is not a choice, and any vendor above it is a choice that the programme can make with the rest of its considerations. Without the floor, procurement becomes a negotiation with an impressive demo.
I do not act as a certifying or accrediting body, and the platform makes no such claim. I am a sole proprietor delivering engineering, evidence gathering and procurement analysis. Formal attestation of a vendor posture is someone else process. What I produce is a documented evaluation with the arithmetic visible and a recommendation driven by the scorecard rather than by the presentation.
The work connects back to ai-model-governance-healthcare for the programme
the vendor feeds, to llm-deployment-healthcare for the deployment the vendor
supports and to healthcare-ai for the overall AI position. Where the vendor
sits alongside a clinical documentation or risk adjustment workflow, the diligence feeds the
same evidence trail that those workflows produce. The vendor is a participant in that
trail, and diligence is how the organisation decides whether the participant is entitled to
be.
What the engagement produces
The deliverable is a written diligence report, a scored scorecard with documents behind each score, a list of recommended contract clauses and a small evaluation on the organisation's own sample. The evaluation harness from the report continues to run after signature as part of the change policy. The report is short — usually fifteen to twenty-five pages — because long diligence reports are unread, and an unread report is worse than no report.
Diligence on a single vendor usually takes two to four weeks, depending on the vendor's responsiveness and whether the pre-signature evaluation is feasible. Diligence on a shortlist of three to five vendors in the same category takes six to eight weeks and produces a comparison that is useful for negotiation and for contract design. The output is not a recommendation in isolation: it is a recommendation with the evidence needed for the rest of the decision.
The engagement starts with a short criteria-setting session with the buyer's clinical, compliance and information-security leadership, so that the diligence is scored against the organisation's criteria rather than mine. From there I run the evidence pack review, the vendor conversation, the pre-signature evaluation and the scorecard. The vendor is treated as a participant in the evaluation, not as a defendant, and the report is written so that the vendor can read it and respond to any finding.
Proof
- ONC (g)(10) self-test PASS, 317 of 317 subtests, April 2026 — public suite URL
- 67 public hostnames, 51 serving, one AWS account
- 104 clearinghouse enrolments across 53 jurisdictions
- USCDI v7 proposal — Patient Digital Identity Credential — listed at isp.healthit.gov
- SAM.gov UEI MVKQW9DKKGB8 · CAGE 20KM5 · NPI 1194645564
- AWS Business Associate Addendum active since 18 August 2026
Where this work is delivered
Engagements run from Las Vegas and reach across the valley: clients in Las Vegas, NV; Las Vegas, Nevada; Henderson; Clark County, Nevada; and North Las Vegas, Nevada — plus remote delivery nationwide.
How an engagement runs
1. Trace one real case
Follow a single patient or claim end to end with a timestamp at every hop, before any document is written.
2. Written findings
A findings list naming the defect, the system, the owner and the date.
3. Monthly review
Review the list monthly and check whether the dates held.
Frequently asked
What is this service?
This page describes charles trinh — ai vendor due diligence — the scope, the failure modes it prevents and how the work is measured. It sits in the Regulatory and Capital Markets category at trinh.io.
Where is the work delivered?
Engagements run from Las Vegas, Nevada across the valley — Las Vegas, Henderson, North Las Vegas, Spring Valley, Paradise, Winchester, Sunrise Manor, Enterprise, Summerlin, Clark County — and remotely anywhere in the United States. On-site work in the Las Vegas valley, remote delivery elsewhere.
What does it cost?
Standard rate $195 per hour, committed $165 per hour at twenty hours or more per week for three months or more; $1,600 standard day, $1,350 committed day. Scoped in writing before work starts, fixed fee where the deliverable is defined.
Who is Charles Trinh?
A technology architect in Las Vegas, Nevada, working in healthcare identity, consent and interoperability. Not a physician, not a medical practice, and not affiliated with any other Charles Trinh.
How does an engagement start?
With one trace of a real case and a written findings list naming the defect, the system, the owner and the date. The first call is twenty minutes and free: 702-559-3489.
Engagement
Scoped in writing before work starts; fixed fee where the deliverable is defined. First call is 20 minutes and free. 702-559-3489 · charles@trinh.io
Talk to Charles
- Phone: 702-559-3489
- Email: charles@trinh.io
- Based in Las Vegas, Nevada · working remotely across the United States
Record
Profile
Healthcare technology architect working where federal regulation becomes production infrastructure. Built and operates a FHIR R4 clinical platform — SMART App Launch authorization server, patient and provider portals, SQLite — on a hardened single-host AWS estate of 67 public hostnames. The platform returned a self-test PASS on the ONC (g)(10) Standardized API test suite in April 2026, 317 of 317 subtests, publicly verifiable. Designed a consent architecture in which a permissioned ledger records patient grant, refusal and revocation as immutable events while all clinical data stays off-chain and encrypted per patient — no PHI on the ledger by design. Filed a proposed USCDI v7 data element with ONC on verifiable patient identity. Translates the Federal Register, HHS/CMS guidance and HITRUST requirements into architecture, controls and audit evidence.
Selected engineering scale
- Self-managed estate of 67 public hostnames, 51 serving, on one AWS account.
- FHIR R4 API with a SMART App Launch authorization server; patient and provider portals; an AI service layer; Hyperledger Fabric peer, orderer and CA; IPFS; a 337-table SQLite clinical store.
- Two permissioned distributed ledgers in production, one operating a consent registry that records grant, refusal and revocation with no PHI on chain.
- X12 005010 pipelines across eight transaction sets.
- Designed, built, deployed and operated end to end by one person.
Credentials and standards work
- Inferno (g)(10) self-test PASS, 317 of 317 subtests, April 2026 — the ONC Standardized API test suite. Standards validated: US Core 6.1.0, USCDI v3, SMART App Launch 2.0.0, Bulk Data 2.0.0. Public suite URL.
- USCDI v7 public comment submission, April 2026 — proposed Patient Digital Identity Credential under the Provenance data class, listed publicly on the ONC Interoperability Standards Platform.
- SAM.gov UEI MVKQW9DKKGB8 · CAGE 20KM5 · NPI 1194645564. AWS Business Associate Addendum active 18 August 2026.
Technical environment
AWS us-east-1 (compute, storage, messaging, SMS, email, voice) · Linux and systemd hardening · nginx · Fabric and EVM ledgers · FHIR R4 · X12 005010 · HL7 · SQLite · self-hosted mail, drive, search and monitoring · Git-based change control and CI release gates.