QlarcPratt capstone · 2025 – 2026

One AI vendor.
One bank.
One blocked deal.

QlarcPratt capstone · 2025 – 2026

I traced it to three gaps, and built a company around closing them.

Role: Business & Product Lead
I owned: Thesis · research · MVP scope · business model
Team: Five, led by me

Scroll

The film

Meet Qlarc.

01My thesis

In finance, AI governance is a legal gate.

A vendor who can't produce governance evidence on demand doesn't get rejected. They never hear back.

89%

of vendors can't see their own AI across teams and third-party services.

eDiscovery Today, 2025

4–6 mo

to produce governance docs that live in engineering, not in auditable formats.

ModelOp, 2024

$2M+

lost per deal when good governance can't be translated into what buyers need.

Primary research, 2026

02Competitive landscape

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

AI-specificGeneral complianceVendor-sideBuyer-sideHolistic AI / ModelOpCredo AIVanta / DrataSecureframeOneTrust / TrustArcResponsive / Inventive AI

qQlarcThe unoccupied gap

03The product

Each step closes one gap.

Closes the visibility gap

Connect

Reads what AI is actually running (model, version, region) without touching customer data.

Evidence PackNorthbank vendor review

Workspace synced19 Mar 2026

Connect your AI stack

Read-only access. Qlarc never touches customer data, prompts, or transactions — only model metadata.

Providers connected

0/4

Integrations

OAuth · revocable anytime

OpenAI

API · model & usage metadata

IDLE

Anthropic

API · model & usage metadata

IDLE

Google AI

API · model & usage metadata

IDLE

AWS billing

CSV · regions only

IDLE

Detected models

0 of 3

gpt-4o

v2024-08-06·us-east-1

claude-3-5-sonnet

v20241022·us-east-1

gemini-1.5-pro

v002·europe-west4

Independently verified from provider API

Closes the documentation gap

Evidence mapping

Every claim is corroborated: the API signal and the document have to agree.

Evidence PackNorthbank vendor review

Workspace synced19 Mar 2026

Evidence mapping

We read your governance docs and link every claim to a procurement question — with a source line you can audit.

Evidence mapped

of 10

Model_Risk_Policy_v3.pdf

extracting · page 1 of 14

1## Model Risk Policy

2version 3.0 · effective 2026-01-15

3Model: gpt-4o · v2024-08-06

41. Model inventory

5Owner: Sarah Chen, Head of Model Risk

72. Bias & fairness testing

8Bias method: SHAP · quarterly cadence

11Override flow: 3-step human review

133. Human oversight

14Decision log retention: 7 yrs

Procurement questions

0/6 on screen

Q01

Model & version governance

SR 11-7

Pending

Q02

Named accountable owner

SR 11-7

Pending

Q03

Bias testing method

ECOA · FCRA

Pending

Q05

Escalation & human override

EU AI Act

Pending

Q06

Output monitoring

FINRA 2026

Pending

Q07

Training data provenance

EU AI Act

Pending

Closes the translation gap

Gap filling

Shows why each gap matters to the buyer, not just what's missing.

Evidence PackNorthbank vendor review

Workspace synced19 Mar 2026

Fill the gaps

Constrained inputs — never free text — with the buyer's reasoning visible in line.

Gaps remaining

of 2

Q04

Bias testing frequency

Missing

ECOAFCRAfrom Regulation Library

How often is bias testing conducted across protected categories?

Frequency

Select frequency…

vendor-provided · strength 0.75

AI Ops reviewer

Why this matters

buyer view

Under ECOA, a credit model with annual-only bias testing is disqualifying. Quarterly cadence with documented action thresholds clears review.

Buyer follow-up

"Attach the most recent quarterly bias report."

Section completeness0%

The human gate

Review & approve

Every answer is editable. Nothing leaves without a named, timestamped sign-off.

Evidence PackNorthbank vendor review

Workspace synced19 Mar 2026

Review & approve

No output leaves until a named human signs. Every approval is timestamped and cryptographically logged.

Sections approved

of 10

Sections

left → right · top → bottom

Q01Model & version governanceAPI

Q02Named accountable ownerAPI

Q03Bias testing methodDOC

Q04Bias testing frequencyGAP

Q05Escalation & human overrideDOC

Q06Output monitoringAPI

Q07Training data provenanceDOC

Q08Data quality & explainabilityDOC

Q09AI component inventoryAPI

Q10Autonomous safeguardsGAP

Vendor approval

non-negotiable

Signed by

Awaiting signature…

Audit record

timestamp 19 Mar 2026 · 14:32 UTC

hash 0x7a3f2cd8b912e041

version Evidence Pack v1.0

Product screens designed by the Qlarc team

04Business case

1 dealpays for itself

One recovered procurement deal covers a year. The ROI case doesn't need a spreadsheet.

45–55%pilot-to-paid conversion

Within 90 days of pilot close: the number that turns a demo into recurring revenue.

$487.5Kyear 2 revenue

25 clients across three pricing tiers, base case.

Year 3break-even

Year 1 builds the platform. Year 2 proves repeatable delivery. Year 3 is the inflection point.

Annual SaaS from $5K to $75K, built as a Public Benefit Corporation.

05The questions every buyer asks

Three risks. Three product decisions.

Data & privacy

Who owns the training data, and where is it stored?

LLM extraction only. No training on customer data.

Signals are read, never retained.

Bias & fairness

Are AI decisions discriminating against protected groups?

Source attribution on every claim, verified by the vendor.

The API signal and the document have to agree.

Accountability

When AI fails, who is responsible?

A mandatory attestation gate before export.

The sign-off is named and timestamped.

My role

I led a team of five as Business & Product Lead: shaping the three-gaps thesis, leading the research, scoping the MVP and building the business model. The product screens were designed by my teammates, Shrutika Rajput, Devanshi Valia, Urvi Karania and Shenao Li.

Business & Product Lead, AI Research · Pratt Institute capstone · May 2025 – April 2026

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 · Typically respond within 24 hours · 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 · Typically respond within 24 hours · 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.

Typically respond within 24 hours · Happy to jump on a quick call

One AI vendor.One bank.One blocked deal.

I traced it to three gaps, and built a company around closing them.

Meet Qlarc.

In finance, AI governance is a legal gate.

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

Each step closes one gap.

Connect

Evidence mapping

Gap filling

Review & approve

Three risks. Three product decisions.

Interested in this work?

Interested in this work?

Interested in this work?

One AI vendor.
One bank.
One blocked deal.