Voice assistants are rapidly evolving from simple single-turn command responders to autonomous agentic copilots. Traditional assistants can set a single timer or check a schedule when explicitly commanded, but the vision for Alexa+ is to handle end-to-end multi-step planning. We built FocusFlow, an agentic workspace copilot that enables Alexa+ to process natural voice queries, retrieve pending tasks, evaluate priorities, and autonomously schedule calendar focus blocks without tedious manual coordination.
What It Does FocusFlow provides an interactive simulated Alexa+ web experience featuring:
Multimodal Voice & Text Input: Users can speak naturally via voice recording or submit text prompts.
Intent-Aware Routing: The agent intelligently distinguishes general conversational greetings from actionable workspace goals.
Autonomous Tool Execution:
get_pending_tasks(): Reads pending tasks and deadline priorities from the workspace.
schedule_focus_block(): Programmatically reserves calendar sessions for critical items.
Real-Time Visual Reasoning: The UI displays live feedback showing exactly which tools are invoked before delivering the final synthesized Alexa+ spoken response.
How We Built It Frontend: Built with Streamlit, featuring separate multimodal tabs for microphone recording and text interaction.
Agentic Engine: Powered by open LLM inference connecting to serverless reasoning models.
Tool Definitions: Custom Python functions structuring task retrieval and calendar event confirmation via JSON schemas.
Workflow Simulation: Replicates the official Alexa+ agentic tool-execution loop to showcase autonomous multi-step execution.
Challenges We Faced API & Model Versioning Transitions: Adapting code to work cleanly across modern SDK updates and finding reliable zero-friction free-tier inference endpoints.
Intent Discrimination: Preventing the agent from aggressively triggering tools during simple pleasantries, resolved by introducing intent routing so tools only fire for actionable productivity requests.
Secret Security: Protecting API credentials by shifting from hardcoded keys to in-memory configuration inputs, ensuring compliance with GitHub push protection.
Accomplishments We're Proud Of Creating an intuitive, responsive voice-and-text interface that faithfully simulates the agentic execution model expected in next-generation Alexa+ integrations.
Achieving an end-to-end workflow where a single conversational command triggers autonomous multi-tool actions.
What We Learned How to structure tool schemas and function-calling prompts for autonomous LLM agents.
Best practices in architecting voice-enabled agentic workflows and managing secure token sessions in client interfaces.
What's Next for FocusFlow Integrating live connections to Google Calendar, Notion, and Jira APIs via a dedicated remote MCP (Model Context Protocol) server.
Adding real-time streaming audio responses using text-to-speech (TTS) synthesis to deliver a fully hands-free voice experience.
Log in or sign up for Devpost to join the conversation.