QlarcPratt capstone · 2025 – 2026

An AI company pitches a bank. The deal goes silent.

Three gates stand in the way. No one tells the vendor.

So we built Qlarc: proof a buyer can approve.

Scroll

Role
Business & Product Lead
I led
Primary research, product strategy, competitive analysis, legal structure
Timeline
One year, May 2025 – April 2026
Outcome
Capstone concept, tested with buyer, vendor and expert interviews

The film

Meet Qlarc.

01The problem

Three gaps, tangled together.

Visibility gapVendors don't know what AI they're using.89%of enterprise AI is invisible to the organizations running it.
Documentation gapNothing is written down or organized.65%of mid-sized vendors have no centralized AI model inventory.
Evidence translation gapWhat exists can't be understood by buyers.$2Min ARR at risk per blocked deal.

The life of a deal

It looks like a straight line. It isn't.

What the vendor seesDeal active in the CRMWhat's really happeningShortlistedQuestionnaire issuedVendorassemblesLegalEngineeringComplianceAI governance reviewBuyer asks for clarificationThird-party dependency surfaces late15–20 days, 40–50 hoursBlocked. No follow-up.Next deal: rebuilt from scratch

02The proof · Primary research

We asked all three sides. They said the same thing.

  1. PP
    Pratik P.Buyer side, AI Ops Leader at a major U.S. financial institution
  2. CM
    Dr. Cari M.Independent expert, Co-Founder of the AI Procurement Lab
  3. DB
    Debashis B.Vendor side, Practice Lead in Technology Consulting
“The vendor never knows it's dead. ”
Pratik P., Buyer side, AI Ops Leader at a major U.S. financial institution

So readiness is checked first.

“Evasive vendors lose regardless of product quality. ”
Dr. Cari M., Independent expert, Co-Founder of the AI Procurement Lab

So gaps are shown, never hidden.

“Documentation gets rebuilt from scratch each cycle. ”
Debashis B., Vendor side, Practice Lead in Technology Consulting

So evidence is stored and reused.

02The proof · Competitive landscape

Every tool was built for a different problem. One corner was empty.

No tool helped vendors produce the evidence buyers are legally required to evaluate. Hover or tap a tool to see why.

AI-specificGeneral complianceVendor-sideBuyer-side

03The product

Each step closes one gap.

The vendor connects, maps, fills and approves. What comes out is an Evidence Pack: buyer-ready answers, each traced to a source.

Step 1Visibility gap

Connect & upload

Reads what AI is running, without touching customer data.

✓Company profile✓Readiness check3Connect and upload
Connect and upload
AI providers
OA
OpenAIgpt-4o
Connected
AN
Anthropic
GE
Gemini
Connect
Documents
Drag files here or browse
Model_Risk_Policy_v3.pdf
Bias_Testing_Report.pdf
Read-only. Never customer data.Continue
Model_Risk_Policy_v3.pdf

Step 2Documentation gap

Evidence mapping

The API signal and the document have to agree.

Meridian AIMR
ProjectsDocumentsIntegrationsRegulationsSettings

Projects

Northbank vendor review Draft
Add documentFill 2 gaps
Evidence mappingGaps 2Review

Verified

Document only

Missing

Questions checked

of 10
Model_Risk_Policy_v3.pdf
1# Model Risk Policy
2Version 3.0, effective 15 Jan 2026
31. Model inventory
4Model: gpt-4o, v2024-08-06Q1
5Owner: Sarah Chen, Head of Model RiskQ2
62. Data
7Training data: licensed financial corpusQ4
8Explainability notes kept per modelQ8
9AI components listed in Appendix AQ9
103. Bias testing
11Method: SHAP, across protected categoriesQ3
12Cadence: quarterlyQ6
136.2 Human review
14Halt conditions, three-step human reviewQ10

Q10 also confirmed by the API

Escalation to a human is on for gpt-4o.

OpenAI usage data, 14 Mar 2026

✓ Both sources agree
QuestionRegulationSourcesStatus
Q1Model and version governanceSR 11-7
Q2Named accountable ownerSR 11-7
Q4Training data provenanceEU AI Act
Q8Data quality and explainabilityECOA, FCRA
Q9AI component inventorySR 11-7
Q10Autonomous safeguardsEU AI Act Art. 14
Q3Bias testing methodECOA, FCRA
Q6Bias testing frequencyECOA, FCRA
Q5Escalation and human overrideSR 11-7
Q7AI output monitoringFINRA 2026

Step 3Evidence translation gap

Gap filling

What a buyer can't approve without, and why.

Meridian AIMR
ProjectsDocumentsIntegrationsRegulationsSettings

Projects

Northbank vendor review Draft
Evidence mappingGaps 2Review
QuestionRegulationOwnerStatus
Q5Escalation path and human overrideSR 11-7, EU AI Act Art. 14
Q7AI output monitoringSR 11-7, FINRA 2026UnassignedMissing
Q5. Who is accountable when your AI makes a wrong decision, and how does it escalate?

Accountable person

Must be a named person. A team name won't pass review.

Escalation steps

Resolution timeframe

Override authority

Where it's documented

How this will appear in the pack

Vendor-provided answer, strength 0.75. Labelled as vendor-provided so the buyer can see where it came from.
Flag as open gapSave and next gap
Why the buyer needs thisWithout a documented escalation path and named owners, the reviewer can't confirm human oversight exists.
Buyer reviewer
AI Ops, model risk
Regulation
SR 11-7, EU AI Act Art. 14
If left open
Shown as a gap in the pack

Found so far

Mentions a three-step review, but names no owner.

Model_Risk_Policy_v3.pdf, section 6.2

Step 4Human gate

Review & approve

Nothing leaves without a named sign-off.

Meridian AIMR
ProjectsDocumentsIntegrationsRegulationsSettings

Projects

Northbank vendor review
Evidence mappingGaps 2Review
QuestionAnswer sourceApproved byStatus
Q1Model and version governanceAPI and document
Q2Named accountable ownerAPI and document
Q4Training data provenanceAPI and document
Q8Data quality and explainabilityAPI and document
Q9AI component inventoryAPI and document
Q10Autonomous safeguardsAPI and document
Q3Bias testing methodDocument
Q6Bias testing frequencyDocument
Q5Escalation and human overrideVendor-provided
Q7AI output monitoringVendor-provided
What safeguards exist when your AI operates without real-time human review?

Drafted answer

Edit

Escalation to a human is on for this model. Halt conditions and a three-step human review are documented, with a named owner at each stage.

Sources: provider API, Model_Risk_Policy_v3.pdf section 6.2

Vendor approval

Signed by

04The market · Go-to-market

Teach first. Sell second.

Starts when a deal gets blocked

Concept mockups of the plan

  1. QlarcAI procurement readiness

    Your AI deal didn't stall. It was blocked, and no one told you.

    Deal in reviewBlocked
    LikeCommentShare
  2. qlarc.ai/readiness-check

    Would your AI pass a procurement review?

    Written AI governance policyYes
    Bias testing reportNo
    Named accountable personYes
    One gap a buyer would flagEmail me my report
  3. The gap a buyer would flagQlarc to you

    Your readiness check found no bias testing report. Under ECOA and FCRA, a reviewer can't approve lending AI without one.

    See how vendors close this in a walkthrough.

    Book a walkthrough
  4. Qlarc

    Welcome. Create your account.

    Work email
    Company
    Deal you're preparing for
    Start your pilotTarget: 45–55% pilot-to-paid within 90 days
  5. FINRA 2026 update2 of your answers need a fresh look.Review
    Renewal readyOnly what changed needs your review.Open
    Target: net retention above 110%
QlarcAI procurement readiness

Your AI deal didn't stall. It was blocked, and no one told you.

Deal in reviewBlocked
LikeCommentShare
qlarc.ai/readiness-check

Would your AI pass a procurement review?

Written AI governance policyYes
Bias testing reportNo
Named accountable personYes
One gap a buyer would flagEmail me my report
The gap a buyer would flagQlarc to you

Your readiness check found no bias testing report. Under ECOA and FCRA, a reviewer can't approve lending AI without one.

See how vendors close this in a walkthrough.

Book a walkthrough
Qlarc

Welcome. Create your account.

Work email
Company
Deal you're preparing for
Start your pilotTarget: 45–55% pilot-to-paid within 90 days
FINRA 2026 update2 of your answers need a fresh look.Review
Renewal readyOnly what changed needs your review.Open
Target: net retention above 110%

04The market · The business

One recovered deal pays for the year.

$750KAt risk per blocked deal
$20KA year for Qlarc
45–55%Pilot-to-paid target
Year 3Projected break-even

Built as a Delaware Public Benefit Corporation, which opens NSF SBIR and CDFI-linked grants.

05How Qlarc builds trust

Remember the vendor with the dead deal? With Qlarc, they walk back in with proof.

AI could invent evidence.Extract only, never generate.
Output mistaken for certification.Document, not certify.
Vendors enter too late.Built for what's coming.

Key learnings

  1. 01

    The market can move your thesis. When the rules behind AI sustainability fell away, we went to where accountability was already enforced.

  2. 02

    Scope is the first design decision. Every expert said go narrow. One sector done well beat covering everything.

  3. 03

    In regulated products, the gate is the feature. A named person slows things down, and that's what makes the output trusted.

Next case studyMuraA class project: a sleep pillow charged by the sun
Let's talk

Interested in this work?

I'd love to walk you through the full case study in a conversation. Reach out and I'll get back to you shortly.

NDA work available on request · helloaroradesigns@gmail.com · Happy to jump on a quick call

Let's talk

Interested in this work?

I'd love to walk you through the full case study in a conversation. Reach out and I'll get back to you shortly.

NDA work available on request · helloaroradesigns@gmail.com · Happy to jump on a quick call

Let's talk

Interested in this work?

I'd love to walk you through the full case study in a conversation. Reach out and I'll get back to you shortly.

helloaroradesigns@gmail.com · Happy to jump on a quick call

© 2026 Radhika Arora, New York
WorkAboutResumeLinkedIn