ScopeAI — Autonomous Competitive Intelligence Agency

Inspiration

Small businesses lose to larger competitors every day — not because their product is worse, but because they can't afford competitive intelligence. Hiring a consulting firm costs $5,000–$15,000. Doing it manually takes 40+ hours per quarter. We built ScopeAI to democratize consulting-grade competitive intelligence for small businesses using autonomous AI agents on Google Cloud.

What it does

ScopeAI is an autonomous competitive intelligence agency powered by three coordinated Gemini 2.5 Flash agents:

1. The Investigator: Uses Gemini 2.5 Flash with live Google Search grounding to extract pricing models, feature matrices, and review sentiment from the live web.
2. The Analyst: Cross-references data against industry context, synthesizing strategic SWOT positioning and unaddressed market gaps.
3. The Writer: Converts findings into an executive-ready Markdown briefing where every material claim links to a verified, clickable web citation.

A customer fills out a brief, pays via Stripe Checkout ($49–$149), and receives a cited report in ~3 minutes while watching real-time agent telemetry.

• What works: The pipeline runs reliably end-to-end, reports contain 31–70 verified live web citations, and the telemetry dashboard provides 100% execution observability.
• What is still in progress: We have zero paying customers to date. The product is live and technically validated on Cloud Run, but early cold outreach generated 0 replies. We are prioritizing organic distribution and public sample briefs over cold messaging.

How we built it

Built solo by a doctor-turned-developer during the hackathon, operating entirely within a €150 Google Cloud promotional credit:

• Gemini 2.5 Flash on Vertex AI with Google Search grounding: Grounding is the core engine. Without real-time web retrieval, competitive reports would just be generic AI hallucinations.
• Google Cloud Run: Serverless containerized deployment with scale-to-zero (min-instances=0) to eliminate idle hosting costs.
• Google Cloud Firestore: Resilient persistence for order states, reports, and streaming telemetry logs.
• Stripe Integration: Signature-verified webhooks triggering the agent pipeline upon checkout completion.
• Next.js 16 + TypeScript & Tailwind CSS: Clean, responsive client dashboard with streaming console logs.

Total spend to date: ~€2.70 (mostly Stripe fees on a refunded founder verification transaction).

Challenges we ran into

1. Grounded JSON Extraction: Gemini occasionally returned empty text when combining complex responseSchema constraints with active Search Grounding. We engineered a multi-tier fallback pipeline: structured schema first, schema-less extraction with regex post-processing second, and standard generation fallback.
2. Cost-Free Judge Sandbox: Evaluating the live demo costs Gemini API credits. We implemented a dedicated sandbox mode with a 15-minute cooldown per IP to let judges test the real pipeline safely without draining credits.
3. Purging Hypothetical Assumptions: Early development mockups included placeholder metrics and hypothetical numbers. We conducted a strict codebase audit to eliminate all unverified claims, anchoring 100% of our narrative in live, measured production telemetry.
4. Cold Distribution Friction: Automated cold outreach to B2B founders yielded zero replies. We learned that selling a new intelligence tool requires visible proof-of-work (public sample reports) rather than unsolicited messages.

Accomplishments that we're proud of

• Autonomous execution: Consistent 96–205s delivery with 31–70 live web citations and zero human intervention in the operational loop.
• Full Observability: A live telemetry console that makes AI decision-making completely transparent to the user.
• Capital Efficiency: Operating an enterprise-grade agent pipeline on Cloud Run + Vertex AI for under €3 lifetime cost.
• Absolute Transparency: 100% evidence-backed submission with zero fabricated metrics or simulated footage.

What we learned

• Citations are the product: Without verifiable live web citations, competitive intelligence is just AI prose. Grounding transforms a language model into an analyst.
• Code is easier than distribution: Building a fully autonomous multi-agent pipeline with payments and telemetry took days; finding the first paying customer takes a tailored go-to-market strategy.
• Engineering honesty wins: Eliminating placeholders and relying solely on measured production telemetry made the entire architecture significantly more robust.

What's next for ScopeAI

The technical feasibility of autonomous competitive intelligence is proven. Our roadmap includes:

1. Acquiring the first 5 paid customers through targeted public sample briefs (e.g., benchmarking HubSpot vs. Salesforce, Notion vs. Confluence).
2. Expanding agency partnerships where marketing and SEO agencies use ScopeAI as a white-label research backend.
3. Automated quarterly monitoring: Triggering scheduled agent re-runs to alert businesses when competitors change pricing or deprecate features.

Built With

• antigravity
• gemini-2.5-flash
• google-cloud-build
• google-cloud-firestore
• google-cloud-run
• google-cloud-vertex-ai
• google-gemini
• google-search-grounding
• next.js
• opencode
• playwright
• stripe
• tailwindcss
• typescript