DevFlow Ai

DevFlow AI is an autonomous agent that detects events, understands intent, and automatically schedules meetings, creates tasks, sends notifications, and tracks workflows

Comment

Inspiration

Every software engineering team suffers from the "context-switching tax." Developers and engineering managers spend up to 40% of their workday manually triaging GitHub issues, copy-pasting meeting links into Google Calendar, updating Jira/Kanban boards, and broadcasting status alerts in Slack.

When we looked at the current landscape of AI tools, we noticed a critical flaw: almost all of them are passive chatbots. They sit in a browser tab and wait for a human to type a prompt.

We asked ourselves: What if an AI didn't wait for prompts? What if it lived in the background, listened to workplace event streams 24/7, and autonomously coordinated schedules, tools, and tasks from start to finish?

That vision became DevFlow AI — an autonomous, event-driven workflow coordinator built to turn real-time software events into immediate, zero-click actions.

What it does

DevFlow AI acts as an autonomous digital chief-of-staff for engineering and product teams:

-24/7 Live Event Ingestion: Continuously listens to incoming Gmail inboxes, GitHub webhooks (issues & PRs), meeting transcripts, and system alerts. -Autonomous Reasoning & Extraction: Uses Google Gemini 3.5 Flash to extract critical metadata (meeting dates, times, participants, issue severities, and actionable tasks) with zero manual tagging. -Zero-Click Google Calendar Cloud Sync: Directly inserts confirmed meetings into Google Calendar via the Google Calendar Cloud API v3, generates deep calendar links (/day/YYYY/M/D), and sets reminder alarms with zero manual clicks. -Interactive Dual-Mode Task Board: Automatically generates organized Kanban cards and list items (Open → In Progress → QA → Completed) attributed with AI priority tags (P0–P3) and audit reasons. Multi-Channel Real-Time Alerts: Dispatches native Windows OS Action Center toasts, browser push notifications, sound chime alerts, and rich Slack webhook messages. -Asynchronous State Machine: Gracefully persists long-running workflows in WAITING states (e.g., waiting for a PR or QA pass) without holding server memory, automatically resuming when follow-up events arrive. -Strict Multi-Tenant Isolation: Every Google account operates in a private, isolated workspace—ensuring User A never sees User B's schedule or data.

How we built it

We built DevFlow AI from the ground up using modern cloud and agentic technologies:

-Agent Framework & Reasoning: Built using the Google Agent Development Kit (ADK) and the Google GenAI SDK (@google/genai) powered by Google Gemini 3.5 Flash for rapid, structured multi-tool decision making. -Calendar & OAuth Integration: Integrated Google APIs SDK (googleapis) for direct Google Calendar Cloud API v3 REST insertions, combined with Google Identity Services (GSI) for multi-tenant OAuth 2.0 authentication. -Email Ingestion Engine: Developed a 24/7 background IMAP streaming service using imap-simple and mailparser with strict privacy exclusion guards (filtering out banking/OTP/marketing emails). -Backend & Persistence: Powered by Node.js (v20+) and Express.js with a Dual-Store Architecture that switches seamlessly between local store (for zero-config evaluation) and Google Cloud Firestore / Google Cloud Pub/Sub. -Frontend Control Surface: Built with React 18 and Vite, styled using a custom Vanilla CSS design system featuring 60fps cinematic animations, optical blur-to-focus scroll reveals, and fluid dark/light themes. -Automated Verification: Comprehensive Jest test suite verifying task models, state machines, and idempotency deduplication.

Challenges we ran into

-Intelligent Idempotency & Deduplication: When streaming high-frequency emails and webhook events, preventing duplicate calendar entries or notification loops was critical. We designed a deterministic idempotency key system based on Message-IDs and payload hashes. -Natural Language Time & Date Normalization: Parsing unstructured date expressions across timezones and informal phrases ("tomorrow at 3pm", "Monday 12 to 3pm") required combining custom regex extractors with Gemini 3.5 Flash structured schemas. -Asynchronous Long-Running Workflows: Traditional server processes cannot stay open waiting days for a GitHub PR to merge. We implemented a decoupled state machine that serializes execution state and wakes up asynchronously upon webhook delivery. -Zero-Click Google Cloud Authentication: Seamlessly handling Google OAuth 2.0 token refreshes on the backend while allowing dynamic client configuration without exposing any secret keys on GitHub.

Accomplishments that we're proud of

-True Zero-Click Autonomy: Seeing an email arrive in a real Gmail inbox and watching DevFlow AI autonomously parse it, create a calendar event, push a desktop toast, and update the Kanban board in under 10 seconds without touching a keyboard. -100% Test Coverage on Core Engine: All unit tests pass cleanly across models, state machines, and deduplication logic. -Bulletproof Multi-Tenant Security: Zero hardcoded credentials in the repository; complete data isolation between authenticated Google accounts. -Cinematic User Experience: An interface that doesn't look like an MVP—complete with 3D optical reveals, responsive glassmorphism, and instant dark/light theme switching.

What we learned

-Agentic vs. Chatbot Paradigm: Event-driven agent architectures that react autonomously to real-world triggers are significantly more impactful for developer productivity than standard conversational chatbots. -Gemini 3.5 Flash Speed: The sub-second latency and structured JSON output capability of Gemini 3.5 Flash makes real-time background event processing viable and cost-effective. -Robust State Persistence: Designing stateful workflows as stateless event consumers makes distributed agent systems scalable and resilient to failures.

What's next for DevFlow Ai

More developer workflows More QA automation Advanced issue and bug triage More integrations Custom workflow templates More event triggers Long-running workflows Human approval checkpoints Better failure recovery Advanced workflow analytics More autonomous multi-step agents

Built With

  • antigravity
  • css3
  • express.js
  • google-agent-development-kit-(adk)
  • google-calendar-api-v3
  • google-cloud
  • google-cloud-firestore
  • google-cloud-pub/sub
  • google-cloud-run
  • google-gemini-3.5-flash
  • google-identity-services-(oauth-2.0)
  • html5
  • imap
  • javascript
  • jest
  • node.js
  • react-18
  • rest-apis
  • slack-webhook-api
  • vite
  • webhooks
Share this project:

Updates

Submission history