Daybridge-Alexa

An Alexa+-inspired AI assistant that turns everyday intentions into structured, actionable plans with real-time execution and progress tracking. Say it. Plan it. Get it done.

Comment

Inspiration

Conversational AI has become incredible at generating text, but largely ineffective at helping users execute real work. When you ask standard voice assistants or chatbots: "Help me prepare for my software engineering interview on Friday. I have one hour each evening," they respond with an overwhelming wall of plain text. The user is left with the friction of manually copying tasks, calculating time budgets, and setting reminders in a separate to-do app. With the Alexa+ initiative, Amazon introduced a paradigm shift: assistants that don't just chat, but take agentic action. We built Daybridge to bring this vision to life. Daybridge bridges the gap between natural language intention and structured software execution. Its core philosophy is simple: "Say it. Plan it. Get it done."

What it does

Daybridge is a simulated Alexa+ web application that converts natural language goals into interactive, trackable daily roadmaps: Intention to Structured Action: Users describe a goal with constraints (e.g., a deadline or daily time budget). Daybridge decomposes the intention into ordered tasks with explicit time estimates, priorities, and deadlines. Context-Aware Follow-Up Mutations: Users can speak or type adjustments like "Make the first day easier and add a mock interview." Rather than starting from scratch, Daybridge references active plan context, adjusts daily workloads, and inserts targeted tasks. Truthful State & Safe Action Pipeline: Borrowing from mission-critical software principles, Daybridge never claims an action succeeded until backend validation passes. Complex changes enter an "Awaiting Confirmation" state with visual diff cards before modifying the database. Interactive Dual-Canvas: Users can view their roadmap in Timeline View (chronological daily schedules) or Board View (focus Kanban cards) with real-time completion tracking and daily capacity meters (e.g., 60 minutes per evening). Multimodal Voice Experience: Built-in Web Speech API voice recognition (microphone input) and text-to-speech synthesis enable users to talk directly to Alexa+ and hear spoken summaries aloud.

How we built it

We designed Daybridge as an integrated monorepo architecture: Frontend: Built with React 19, TypeScript, and Vite. Styled using a custom 3-tier surface depth design system based on a cool neutral palette (Gray 0–950) with an energetic Harbor Teal (#0F6F6B) action accent and Lucide iconography. Backend Brain: Developed with Python 3.13 and Flask. It features a modular service layer orchestrating the Google Gemini API (gemini-3-flash-preview) with structured JSON schema outputs and whitelisted function tools (create_plan, propose_plan_update, list_tasks, answer). Database & Persistence: SQLAlchemy ORM supporting PostgreSQL with an automatic, resilient zero-downtime SQLite fallback (daybridge.db) for consistent local development and cloud availability. Trace & Telemetry: Every request emits a 4-stage operational pipeline trace:
Cloud Infrastructure: The frontend is deployed globally on Vercel, and the backend is containerized with Gunicorn on Render.

Challenges we ran into

Strict Structured Output from LLMs: Ensuring the generative AI strictly produced validated schema payloads without hallucinations or markdown formatting errors. We implemented strict Pydantic schemas and a resilient JSON extractor with retry backoffs. State Mutability vs. Hallucinated Deletions: When a user asked to "Make the first day easier and add a mock interview," naive prompts frequently deleted unrelated tasks. We resolved this by building a dedicated ProposedAction pipeline that models changes as discrete diff records (changed, added, removed) before mutating state. Cross-Platform Audio & Microphones: Browser autoplay policies and inconsistent speech synthesis voices across different operating systems required dynamic voice selection heuristics and graceful audio fallback triggers. 48-Hour Rapid Integration: Aligning the frontend design system primitives with the backend REST endpoints under a tight hackathon deadline. We solved this by publishing an enforceable API_CONTRACT.md before merging the branches.

Accomplishments that we're proud of

End-to-End Golden Demo: Delivering a seamless flow where users speak an intention, see structured tasks appear in sub-second latency, modify them conversationally, and interactively track progress. Zero Fluff / Zero Hardcoding: Every single plan, recommendation, and follow-up adjustment is dynamically generated and persisted in a live database. Truthful State Reporting: Eliminating the deceptive "Your task has been completed" chatbot antipattern by implementing explicit confirmation stages and real backend status checks. Clean Code & Monorepo Separation: Successfully shipping a full-stack, enterprise-grade codebase with 100% test coverage on our primary user journey

What we learned

Prompt Engineering is Tool Engineering: High-performing agentic assistants depend less on lengthy text instructions and more on well-defined JSON action contracts and granular tool constraints. The Power of Visible Pipeline Latency: Showing judges real operational stages (Heard ➔ Understood ➔ Planning ➔ Ready) demystifies AI actions and builds immense user trust. Resilient Architecture Pays Off: Designing the database layer to automatically fall back to SQLite when local database services were stopped ensured our team never lost a minute of development time.

What's next for Daybridge-Alexa

Native Alexa Skill & MCP Server: Implementing a Streamable HTTP Model Context Protocol (MCP) server conforming to the official Alexa+ 2026 specification for real Echo device integration. 1-Click .ics Calendar Sync: Two-way synchronization with Google Calendar, Apple Calendar, and Microsoft Outlook. Proactive Rebalancing: Intelligent rescheduling algorithms that detect when a user falls behind on evening milestones and automatically cascade uncompleted tasks across the week.

Built With

Share this project:

Updates

Submission history