Inspiration
Operations teams waste countless hours manually parsing messy invoices, cross-referencing logs, and coordinating follow-up emails. We built TaskFlow AI to solve this bottleneck by acting as an autonomous operations partner that executes multi-step workflows from a single natural language prompt.
What it does
TaskFlow AI is an autonomous taskmaster agent built using the Google Agent Development Kit (ADK) and Gemini. When a user inputs an unstructured invoice or task request, the agent:
- Intelligently parses the natural language intent.
- Automatically triggers custom Python backend tools.
- Simultaneously logs structured transaction data into a local database file and drafts confirmation emails.
- Returns a clean, verified confirmation directly through an interactive Streamlit web interface.
How we built it
- Agent Framework: Google Agent Development Kit (ADK) & Gemini for function calling and intent parsing.
- Backend Tools: Custom Python operations scripts for database logging and mock email dispatching.
- Frontend & UI: Streamlit for a responsive, clean user experience.
- Cloud & Deployment: Hosted live on Streamlit Cloud with secure API credential management.
Challenges we ran into
Handling asynchronous multi-tool execution and ensuring reliable function-calling outputs from unstructured text required fine-tuning the agent's system prompts and structuring clean tool declarations using Google ADK.
Accomplishments that we're proud of
- Successfully built a multi-tool autonomous agent from scratch.
- Deployed a fully functional, cloud-hosted version live on the internet during the hackathon timeline.
- Designed a clean separation of concerns between the Gemini agent brain, Python tools, and the Streamlit frontend.
What we learned
We gained deep experience leveraging Google's agentic tooling, managing tool-use schemas, and rapid-prototyping cloud-native AI applications.
What's next for TaskFlow AI
- Integrating live email APIs (like Gmail/SendGrid) for production dispatching.
- Scaling to enterprise databases (PostgreSQL/Firestore) for team-wide usage.
Log in or sign up for Devpost to join the conversation.