StartRight Launch Desk — Build with Gemini XPRIZE Evidence

Entrant: QuirkWorks LLC

Product: StartRight Launch Desk

Challenge: Build with Gemini XPRIZE

StartRight provides business filing guidance, not legal, tax, or accounting advice. Requirements vary by jurisdiction and change over time. Always verify with official government sources or qualified professionals. StartRight does not guarantee that any filing list is complete.

What StartRight does

StartRight helps founders understand the business filings they may need before they operate. It turns scattered federal, state, city, county, tax, DBA, license, employee, industry, and ongoing compliance requirements into:

1. A free starter filing-readiness checklist
2. A $49 Launch Packet Review (risk review + prioritized steps + verification checklist + 7-day action plan)
3. A $149 Concierge Launch Plan
4. A $299 Multi-Entity Compliance Map

Which categories it fits

• Entrepreneurship and small business services — tooling that helps new founders launch correctly.
• Professional services access — makes filing-readiness guidance affordable and self-serve.
• Compliance — organizes regulatory/filing obligations.
• AI agents / founder tools — the business is operated by autonomous Gemini agents.

(These map to the seeded contest_reuse tags: entrepreneurship, small_business_services, professional_services_access, compliance, AI_agents, founder_tools.)

How Gemini is used

All AI is performed server-side only through a single runtime (lib/agents/runtime.ts) that calls the Google Generative Language API (Gemini) with a key from the environment. Six agents:

Every agent returns structured JSON and `suggestedTags`. If GEMINI_API_KEY is missing, each agent uses a safe deterministic fallback so the business keeps running; the run is still logged with status fallback.

Which Google Cloud product is used

Google AI / Gemini API (Generative Language API), model gemini-1.5-flash by default (configurable via GEMINI_MODEL). Usage is explicit and logged on every call: model name + Google Cloud product are written to the AgentRun record and to a structured CloudWatch log line (evt: "startright.agent_run").

How AWS is used

• Amazon Cognito — user accounts, login/logout, admin group. (amplify/auth/resource.ts)
• Amazon DynamoDB — all application data via Amplify Gen 2 data models. (amplify/data/resource.ts, single-table access in lib/db.ts)
• AWS Lambda / serverless functions — agent execution, Stripe webhook, and API routes run as serverless functions (Netlify Functions in the reference deploy; AWS Lambda via Amplify is the documented native path).
• Amazon SES — transactional email; drafts are persisted when SES is not configured. (lib/email.ts)
• Amazon CloudWatch — every agent run, payment, email, and webhook emits a structured JSON log line captured by the function runtime / CloudWatch.
• Amazon S3 — available for evidence exports if needed.

The app ships dual backends: AWS-native when configured (DynamoDB/Cognito/SES), and a local fallback (JSON store + local auth + email drafts) so it runs and builds with zero credentials. This keeps the project deployable and demonstrable at all times.

How AI agents operate the business

The critical revenue path is fully agent-operated:

Visitor → founder intake → Gemini Checklist Agent → preview → account (Cognito)
→ saved checklist → $49 purchase (Stripe) → Gemini Risk Review + Packet Writer
→ paid review in dashboard → SES follow-up email → admin/evidence metrics + CloudWatch logs

No human writes checklists, reviews, qualification, or emails — Gemini agents do, with full audit logging.

How users are tracked

• UserProfile (Cognito-backed) with roles guest / customer / admin.
• Guest previews are captured by leadEmail and claimed to the account on signup.
• Admin can view users, business profiles, checklists, checklist items, payments, reviews, agent runs, feedback, and tags.

How revenue is tracked

• Payment records (Stripe session/intent, product, amount, status). Stripe Checkout for all paid products; a manual_pending fallback logs intended purchases when keys are absent.
• Revenue, conversion rate, and revenue-by-product are computed in lib/services/metrics.ts and shown on /admin/metrics and /evidence.
• Revenue events are tagged paid_review, revenue_related.

How agent logs are tracked

• Each call writes an AgentRun (agent name, model, Google Cloud product, input/output summary, status, revenueRelated, AWS execution context, error).
• Each call emits a structured CloudWatch log line and is tagged with XPRIZE evidence tags: gemini_agent_run, google_cloud_usage, aws_usage, cloudwatch_logged, and revenue_related where applicable.

How customer evidence is collected

• CustomerFeedback (rating, comment, permission-to-anonymize) tagged customer_evidence, testimonial_candidate, traction_evidence.
• The Evidence Agent produces an anonymized summary (no names/emails) for submission.

First-class tagging

Tags are real records (Tag + TagAssignment), never comma-separated strings. Nine targetTypes and four sources (system, gemini, admin, user). Admin can filter records by tag. Customers only ever see a small, non-alarming subset (Federal, State, City, County, Tax, License, DBA, Needs verification).

Pre-existing boilerplate / templates

This codebase was built from scratch for StartRight. There is no pre-existing third-party boilerplate beyond the standard Next.js App Router scaffold and open-source libraries (Next.js, React, Tailwind, Stripe SDK, AWS SDK, Zod). The deterministic agent fallbacks are original and used only when a live Gemini key is absent.