Existing project Full-stack

Building an AI Productivity Assistant for ADHD

Engineering a conversational AI platform for ADHD-focused productivity, with streaming responses, persistent conversations, and AI-powered task actions.

Role
Fullstack Engineer
Year
2024 · Archived
Stack
Python (FastAPI), Svelte
How comigo streaming and persistence work
Streaming responses and conversation persistence.
How comigo natural language task execution works
Natural-language task execution.

What I built

ComigoAI is an AI-powered productivity assistant built to support people with ADHD in managing tasks, conversations, and daily challenges. The product combines conversational AI with practical productivity workflows, allowing users to interact with an AI assistant directly through a browser extension.

I contributed to the development of the platform across the browser extension and backend, working on the conversational architecture, streaming AI responses, conversation persistence, and task-oriented interactions. The platform was designed to go beyond a traditional chatbot by connecting natural-language conversations with actionable productivity features.

Comigo also included a daily reflection experience where users could describe how their day was going and receive AI-assisted feedback based on their responses.

Browser extension

  • A Svelte-based browser extension where users talk to the assistant directly
  • AI responses shown as they stream in, rather than after the complete response is ready
  • A daily reflection experience where users describe their day and receive AI-assisted feedback

Backend

  • A Python/FastAPI backend acting as the application and AI integration layer
  • Conversations persisted in the database so earlier interactions stay available as history
  • Natural-language requests interpreted and executed as task operations, such as creating a task

How a message flows

  1. Ask

    The user writes to the assistant in the browser extension.

  2. Stream

    The backend streams the AI response back to the browser as it is generated.

  3. Persist

    The conversation is stored in the database and stays available as history.

  4. Act

    When the message asks for a task, the backend interprets it and runs the task operation.

Outcome

  • Established a conversational AI architecture combining real-time streaming with persistent conversation history.
  • Enabled users to perform productivity actions through natural language.
  • Connected AI conversations with application functionality rather than limiting the assistant to informational responses.
  • Created a daily reflection workflow that extended the assistant beyond traditional productivity interactions.
  • Built the foundation for an AI-powered assistant that could combine conversation, context, and application actions within a single interface.

Want something like this built?

Get in touch