LaunchQA

LaunchQA is a launch-readiness workbench for customer-specific AI website agents.

LaunchQA tests whether an agent built from messy customer docs, FAQs, decks, product pages, and sales notes is ready before it touches real visitors. The product insight is that adaptive AI website agents are only useful if each customer deployment can be tested for coverage, grounding, routing, and handoff quality before launch. The hard part is not the avatar; the hard part is knowing whether the agent has enough customer truth to answer safely, route buyers to the right content, extract useful intent, and flag launch blockers. The MVP demonstrates that loop with a synthetic customer, Northstar Compliance, and an intentional missing HubSpot integration FAQ.

Independent portfolio prototype. Not affiliated with Interact AI. Uses synthetic customer data only.

Hero Loop

Missing HubSpot coverage
  -> buyer scenario fails
  -> LaunchQA marks it as a launch blocker
  -> add missing FAQ
  -> re-run
  -> grounded answer passes and readiness improves

Why This Exists

AI website agents can adapt to a visitor, but customer deployments still need a readiness layer before real traffic touches the agent. Each customer brings different product pages, security notes, FAQs, pitch decks, pricing caveats, sales notes, and integration claims.

For Interact-style AI website agents, the visible experience is only one part of launch. The deployment team also needs to know whether customer-specific product knowledge is complete, whether answers are grounded in source material, whether the right content surface is selected, whether buyer intent is captured, whether CRM handoff fields are useful, and whether any missing source material should block launch.

LaunchQA focuses on that pre-launch workbench.

Hero Demo

1. Seed Northstar Compliance.
2. Open the Knowledge Map and show that HubSpot is missing from source coverage.
3. Run buyer scenarios.
4. Open the Report and show HubSpot as a launch blocker.
5. Open the HubSpot run detail and show missing_source_coverage.
6. Add the HubSpot FAQ and re-run.
7. Show HubSpot passing with citations and readiness improving.

What Is Real

• Deterministic seeded customer docs
• Knowledge map
• Scenario-based readiness runs
• Grounded answer checks
• Unsupported claim and missing source coverage checks
• Content surface routing
• Buyer intent extraction
• CRM-shaped handoff
• Launch readiness report
• Automated tests

What Is Simplified

• Local JSON store for deterministic demo storage
• No avatar
• No voice
• No real CRM OAuth
• No real calendar booking
• No external integrations
• Scenario-based staging agent, not an arbitrary production chatbot
• Synthetic customer data only

Screenshots

Regenerate these with npm run screenshots while the local app is running.

Home page: the README-first framing and hero loop for customer-agent launch readiness.

Import page: seeded synthetic Northstar Compliance material and the intentional HubSpot gap.

Knowledge map before FAQ: source coverage is visible before the missing HubSpot content is added.

Scenarios page before run: buyer-readiness tests are ready to execute.

Report with blocker: HubSpot fails because imported source coverage is missing.

HubSpot run detail: missing source coverage, insufficient evidence, routing, intent, handoff, and eval checks.

Report after FAQ re-run: the HubSpot FAQ adds citations and readiness improves.

Run Locally

npm install
npm run db:push
npm run seed
npm run dev

Open http://localhost:3000.

Verification:

npm run test
npm run typecheck
npm run build
npm run verify

Tech Stack

• Next.js App Router
• React
• TypeScript
• Local JSON store
• Deterministic retrieval and eval logic
• Vitest domain tests

Deployment Note

LaunchQA uses a local JSON store for deterministic MVP storage. For a hosted multi-user deployment, the storage layer should move to Postgres/Supabase or another persistent database.

Why This Is Not An Interact Clone

This prototype does not build an avatar or recreate the visible website agent. It focuses on the deployment-readiness layer before a customer-specific agent goes live.

How This Maps To Customer-Agent Launch Readiness

LaunchQA answers the questions a deployment engineer, founder, or Head of Engineering should ask before launch:

• Does the customer knowledge base cover the buyer question?
• Can the staged agent answer with citations?
• Does the agent avoid unsupported integration, compliance, and pricing claims?
• Did it route the buyer to the correct card, page, or slide?
• Did it extract buyer role, pain points, requested integrations, objections, urgency, and buying stage?
• Is the CRM handoff structured enough for sales follow-up?
• Which content gaps block launch, and what source material fixes them?

Known Limits

LaunchQA is a practical deterministic MVP, not production infrastructure. It does not include multi-user auth, real file ingestion, embeddings, LLM judging, real CRM field mapping, or hosted persistent storage.