Inspiration
As students, balancing lectures, group projects, assignments, and studying for exams quickly becomes overwhelming. Valuable ideas and important assignment details discussed during casual chats after class or while walking across campus often get forgotten before anyone has time to sit down and write them in a planner.
When we move into the workplace, this problem only grows. Developers lose hours every day doing boring administrative work: typing out meeting notes, creating task tickets, and writing daily status updates.
We were inspired by wearable AI technology like Bee, which listens quietly in the background to capture daily conversations. We thought: Why should students or developers stop what they are doing just to type out notes?
We created BceeLi to bridge this gap. Whether you are a student talking through a class project or a developer planning a system architecture, BceeLi turns your everyday spoken conversations into organized notes, study tasks, and work logs automatically. It gives you time back to focus on what actually matters—learning and building.
What It Does
BceeLi acts as an intelligent, hands-free assistant that bridges wearable ambient data with student and developer workflows:
- Automatic Summaries & Dev Logs: Parses everyday conversations and technical discussions into clean, structured Markdown daily updates.
- Smart Backlog & Task Generation: Automatically converts spoken ideas, assignment details, and bug mentions into organized task items with assigned priority levels.
- Contextual Focus Control: Evaluates focus metrics to help manage distraction levels during study blocks or deep coding sessions.
How We Built It
We designed BceeLi with a lightweight architecture centered around Go and the Model Context Protocol (MCP) to ensure fast processing and low setup overhead:
- Data Ingestion Layer: We configured a pipeline to process real-time audio transcripts and device telemetry, structured as JSON event payloads simulating Bee CLI data feeds.
- MCP Server: Built a custom MCP server using Go to define clear agent actions (such as generating daily logs, parsing task items, and updating workspace state).
- LLM Orchestration: Integrated an LLM agent that receives the wearable context via the MCP server, extracts key intent, and runs the corresponding tool calls.
To determine task priority from incoming audio transcripts, we use a simple scoring formula:
$$S = w_1 \cdot C + w_2 \cdot F$$
Where (S) represents the total priority score, (C) is the transcript confidence ((0 \le C \le 1)), (F) is the user's focus score ((0 \le F \le 1)), and (w_1, w_2) are weighting factors balancing context against active focus.
Challenges We Ran Into
- Filtering Noise from Intent: Ambient audio captures casual background banter alongside important ideas. Tuning prompt schemas to filter out everyday noise while keeping actual tasks was a major focus.
- Testing Without Hardware Dependencies: Building a wearable-first experience without requiring physical dev boards meant designing realistic local JSON data pipelines to test different user scenarios reliably.
- Response Latency: Ensuring that continuous audio processing translated into near-instant workspace actions without noticeable lag.
Accomplishments That We're Proud Of
- Zero-Friction Context Capture: Successfully turning passive spoken sentences into structured Markdown files and actionable project items with zero manual typing.
- Lightweight Go Architecture: Delivering a fast single-binary MCP server in Go that processes incoming event feeds instantly with minimal memory overhead.
- Dual-Purpose Design: Creating a solution that solves real daily problems for students in school while scaling naturally into professional developer workflows.
What We Learned
Building BceeLi highlighted how powerful the Model Context Protocol (MCP) is for connecting LLMs to real-world context feeds. We learned that the best productivity tools are ambient—they run quietly in the background, taking care of administrative tasks so users can stay focused on high-value work.
What's Next for BceeLi
- Live Bee CLI Hardware Integration: Connecting the backend directly to physical Bee hardware and Apple Watch stream feeds.
- Multi-Platform Task Exports: Adding one-click export actions to syncing tools like Notion, GitHub Issues, Jira, and Google Calendar.
- Personalized Context Memory: Expanding the agent's memory model so it recognizes recurring project names, course codes, and team members automatically over time.
Log in or sign up for Devpost to join the conversation.