FlowClose

💡 Inspiration

Software agencies, enterprise consulting teams, and freelancers lose over $42 Billion annually to silent scope creep, ambiguous client briefs, and unending negotiation loops.

Client messages like "Can we also add WhatsApp voice transcription before Friday?" seem harmless to clients but silently add dozens of unbilled engineering hours and break launch dates.

We built FlowClose to automate the entire project lifecycle with autonomous agent intelligence: transforming chaotic client briefs into cryptographically locked specifications, intercepting out-of-scope requests in real time, and dispatching multi-cloud pipelines across Google Cloud and Azure.

🚀 What FlowClose Does

FlowClose is an autonomous, event-driven multi-cloud delivery agent engineered for The Taskmaster track:

1. Autonomous Intake & Spec Synthesis: Takes unstructured client briefs and runs them through Gemini 3.7 Flash to extract concrete deliverables, define constraints, and generate verifiable milestones.
2. Cryptographic Baseline Locking: Generates a full IEEE-standard Software Requirements Specification (SRS) locked with a SHA-256 hash, establishing a single unalterable source of truth.
3. Live Scope Sentinel: Intercepts client messages during active sprints, compares requests against the baseline specification in Google Cloud Storage, detects scope deviation risks, and calculates budget/timeline impacts.
4. Autonomous Change Order Drafting: Generates formal Change Orders with compromise options when scope expansions occur.
5. Human-In-The-Loop (HITL) Comms Queue: Routes drafts through a PM review portal before dispatching client notifications via Azure Logic Apps.
6. Multi-Cloud Event Mesh: Executes asynchronous workflows in <800ms across Google Cloud Pub/Sub, Cloud Run, Cloud Storage, Vertex AI, Azure Logic Apps, and Azure Service Bus.

🛠️ How We Built It

• Intelligence Core: Google Gemini 3.7 Flash & @google/genai SDK for brief synthesis, semantic diffing, and multi-persona advisory.
• Google Cloud Infrastructure:
  • Google Cloud Run: Containerized backend API and agent orchestrator.
  • Google Cloud Pub/Sub: Asynchronous event mesh topics (client-intake, scope-deviations, audit-trail).
  • Google Cloud Storage: Immutable storage for baseline IEEE SRS specifications and SHA-256 hashes.
  • Google Cloud Logging: Audit trail for all autonomous agent decisions.
• Multi-Cloud Azure Services:
  • Azure Logic Apps & Service Bus: Automated communication dispatch and client notification workflows.
  • Azure Blob Storage: Redundant cross-cloud artifact replication.
• Frontend & UI: React 19, TypeScript, Vite, Tailwind CSS, Lucide icons, and Motion design.

🧗 Challenges We Overcame

1. Accurate Scope Creep Detection: Calibrating prompts and system instructions so the agent can distinguish between a harmless clarification and a high-risk scope change.
2. Sub-Second Multi-Cloud Orchestration: Building a unified event bridge between Google Cloud Pub/Sub and Azure Logic Apps with resilient fallbacks and full telemetry.
3. Safe Agentic Autonomy: Enforcing a strict Human-In-The-Loop gatekeeper pattern to guarantee no unverified AI email ever reaches a client.

🏆 Accomplishments We're Proud Of

• End-to-end multi-cloud pipeline executing 6 stages in under 750 milliseconds.
• Mathematical SHA-256 cryptographic verification of project baselines.
• Fully functional live deployment on Netlify with real-time Gemini API integration.
• Seamless agent personas (Lead Architect, Scope Sentinel, Multi-Cloud DevOps).

🔮 What's Next for FlowClose

• Native Jira, Linear, and GitHub Issues two-way synchronization.
• Automated Stripe escrow milestone releases upon acceptance criteria sign-off.
• Multi-agent autonomous PR code review against the locked IEEE SRS.

Built With

• azure
• azure-logic-apps
• gemini-api
• google-cloud
• google-cloud-run
• google-pubsub
• node.js
• python
• react
• rest-api
• typescript
• vertex-ai
• vite