neuroflow - 26

AI study companion built for neurodivergent students. 4 agents collaborate in real-time, researching strategies for your disability, syncing coursework, and scheduling around your life.

Comment

Inspiration

15–20% of college students are neurodivergent (ADHD, autism, dyslexia) and the support system is broken. When a student needs help, they book a meeting with a disability advisor. Then they wait. The deadline doesn't. Generic AI tools give every student the same answer. They don't know your disability. They don't know your schedule. They don't know your coursework. They don't know you. We built NeuroFlow because these students don't need another chatbot, but need AI that adapts to their brain and acts right now.

What it does

NeuroFlow is a multi-agent study companion where 4 AI agents collaborate in real-time to support neurodivergent students. You start by setting up your profile — disability type, preferred session length, peak focus time, and personal challenges. From there, every interaction is personalized:

Then, ask something like "Help me study for my AI midterm" and watch the agents work: the Advisor agent researches evidence-based study strategies specific to your topic and disability via DuckDuckGo, pulls your assignments and grades from Canvas LMS, then passes its findings to the Focus agent. Focus designs a session plan optimized for your attention span. The Calendar agent scans your Google Calendar, finds gaps between your actual classes, and proposes study slots with contextual reasons like "Right after CS180 Lecture, review while material is fresh." Or say "Plan my day" and the system structures your entire day, classes, gym, meals, study blocks, around your neurodivergent needs.

How we built it

We built NeuroFlow on Fetch.ai's uAgents framework with all 4 agents registered and active on Agentverse:

Orchestrator — Intent classification via ASI:One Mini, routes queries to the right agent chain, formats final responses with markdown Advisor — Stores disability profiles in persistent storage, runs live web research via DuckDuckGo tailored to the student's topic + disability + preferences, synthesizes actionable advice through ASI:One Mini Focus — Builds personalized session plans based on Advisor's research, optimizes session length and break intervals for the student's specific disability Calendar — Google Calendar MCP integration for reading/writing events, finds free blocks, proposes study slots with schedule-aware reasoning that references actual classes The agents communicate via a shared state chain: Orchestrator → Advisor → Focus → Calendar → Orchestrator, with data flowing through SharedAgentState.chain_data. For the demo, we built a React 19 + Vite frontend with a FastAPI backend that streams each agent step via NDJSON, so you can watch the collaboration happen in real-time. We also integrated Canvas LMS for assignments/grades and Google Calendar for real schedule management.

Challenges we ran into

Getting the agents to reason intelligently about the student's actual situation was harder than expected. Early versions would suggest studying for an unrelated assignment when the student asked about their AI midterm- the LLM had no concept of relevance. We had to build cross-referencing logic so the system connects Canvas assignments to the student's actual query, matches calendar events to course schedules, and constructs research queries specific to the topic rather than generic disability searches.

Accomplishments that we're proud of

We are proud of how the system is deeply tailored to each individual: your specific disability, your preferred session length, your peak focus time, your actual coursework and deadlines. But at the same time, the underlying architecture works broadly across ADHD, autism, dyslexia, and other neurodivergent profiles. The same agent chain adapts its research queries, session plans, scheduling logic, and even its tone based on whoever is using it.

What we learned

We learned how to build and deploy a multi-agent system using Fetch.ai's uAgents framework, registering agents on Agentverse, chaining them through shared state, and streaming their interactions to the frontend in real-time. Orchestrating multiple data sources (web research, Canvas LMS, Google Calendar) into a single coherent pipeline was a new challenge that sharpened our understanding of agent-to-agent communication. Most importantly, we learned how every detail matters, tone, timing, and reasoning all need to feel intentional.

What's next for Neuroflow

We plan to add an Accommodation Agent that helps students draft disability accommodation requests and tracks university filing deadlines, a real pain point no one else solves. Long term, we envision adding voice interaction and expanding to support K-12 students and workplace neurodivergent professionals.

Built With

Share this project:

Updates