Inbox2Action: Enterprise Taskmaster Agent

• 

• 

Inspiration

Enterprise teams lose hundreds of hours every month manually triaging operational noise—reading lengthy support emails, parsing bug reports, scheduling follow-ups, and copying data between systems. Static chatbots require step-by-step human prompts to be useful. We wanted to build a true Taskmaster Agent that takes raw input, evaluates context autonomously, extracts key parameters, formulates execution plans, and generates ready-to-send responses in seconds.

What it does

Inbound Task Ingestion: Ingests raw customer emails, bug reports, and invoice requests through a unified command center.

Autonomous Parameter Extraction: Parses task priority (High/Medium/Low), category, core action items, and target deadlines using gemini-2.5-flash.

Execution Plan Generation: Formulates a step-by-step resolution strategy without requiring continuous human guidance.

Automated Response Drafting: Drafts professional, context-aware follow-up emails ready for immediate dispatch.

Real-time Telemetry & Analytics: Displays hourly throughput load charts and radar-trace confidence scores across policy, extraction, and drafting parameters.

How we built it

AI Engine: Powered by Google's google-genai SDK and the gemini-2.5-flash model.

Frontend Command Center: Built with Streamlit and enhanced with custom CSS to deliver a clean enterprise operating dashboard.

Data Visualization: Built interactive throughput and execution telemetry charts using Pandas and Plotly.

Environment & Security: Handled credentials via python-dotenv for local execution and secure environment variables in production.

Cloud Infrastructure: Fully containerized using Docker for seamless deployment on Google Cloud Run.

Challenges we ran into

Output Consistency: Crafting precise system instructions so the model consistently extracts structured fields alongside human-readable action plans across varied input types (e.g., invoices vs. bug reports).

UI State & Performance: Integrating interactive Plotly telemetry charts with Streamlit state management without causing page reload lag during agent processing.

Deployment Optimization: Configuring Docker containerization to keep the container lightweight while ensuring all data science and Google SDK dependencies installed cleanly for Google Cloud Run compatibility.

Accomplishments that we're proud of

Transformed a simple prompt interface into a full-scale Enterprise Taskmaster Command Center.

Achieved sub-2-second contextual extraction, plan generation, and response drafting.

Built a fully reproducible repository equipped with Docker and Google Cloud Run deployment pipelines.

Collaborated effectively across specialized roles (AI Engineering, Data Engineering, Google Cloud, Data Science, and MLOps).

What we learned

Structured system prompts within the google-genai SDK dramatically increase the reliability of agentic workflows in enterprise applications.

Coupling LLM reasoning with real-time operational telemetry builds significantly more trust for users evaluating autonomous decision-making.

What's next for Inbox2Action: Enterprise Taskmaster Agent

Google ADK & Multi-Agent Orchestration: Expand the system into a multi-agent framework where specialized sub-agents handle specific database writes and external API calls.

Direct Webhook Integrations: Connect live Gmail, Slack, and Jira webhooks for real-time background task ingestion.

Google Cloud Infrastructure Expansion: Utilize Firestore for persistent cross-session memory banks and Pub/Sub for high-throughput asynchronous message queues.

Built With

• gemini
• streamlit