BrainStormX

Intelligent Brainstorming and Collaborative-Innovation Platform

Comment

BrainStormX Project Story

From Inspiration to outcomes: A visionary tale about building the future of collaborative intelligence.

Inspiration - Where it all began.

Every creative leader knows the frustration when it comes to generating fresh new ideas under pressure. Yet, the very workshops designed to spark innovation, drains us not only in participation, but also in preparation, organization, facilitation and post follow-ups.

Hours are spent preparing agendas, researching context, planning activities, moderating debates, ensuring the group stays focused on objectives, and manually compiling notes that no one ever reviews later.

The very energy that should fuel creativity is lost in logistics.

That pain became our motivation. We envisioned a world where AI orchestrates the messy, and repetitive parts of collaborative-innovation, creating a reality that frees humans to focus on imagination and creativity while being connected.

Our goal was bold yet clear:

To turn chaotic workshops into self-running, context-aware ecosystems that learn and adapt to the culture and values of every team.

That vision became BrainStormX, a full-fledged, autonomous AI-native workshop facilitator built on Amazon Bedrock.

What it does

BrainStormX is an autonomous AI-driven workshop facilitator that leverages the reasoning capabilities of Amazon Bedrock Nova models to transform brainstorming chaos into structured, outcome-driven collaboration. It plans, moderates, facilitates, analyses, provides helpful guidance, and even summarizes and documents workshop sessions as they progress. Facilitators can focus on people and creativity while AI handles planning, orchestration, and follow-up end-to-end.

BrainStormX begins every workshop from context, not from scratch, before the session ends, the action plan is ready, and even before participants leave the room, execution has already begun. That is the BrainStormX promise.

How We Built It

The Agentic Core

At the heart of BrainStormX is a multi-agentic AI facilitation loop embedded with Amazon Bedrock Nova Lite and Nova Pro:

  • Nova Lite manages real-time facilitation and participant engagement.
  • Nova Pro handles complex analytical tasks such as feasibility analysis, agenda framing, action planning, and report generation.
  • Nova Canvas (Titan Image) enhances participant profile photos for consistent visual representation in the workshop.
  • Amazon Bedrock AgentCore Memory provides persistent context recall across sessions.

Workshop Orchestration

We orchestrate the workshop through three reasoning phases;

  1. Plan - Nova Pro analyzes the workshop’s context, goals, and prior sessions to generate facilitator strategies and prepare the session briefing.
  2. Act - The AgentCore Gateway executes declarative actions such as posting ideas, clustering similar themes, running feasibility analysis, and generating action plans. Each phase’s output feeds into the next.
  3. Compose - Nova Lite synthesizes human-like facilitator guidance, generating summaries, next-step recommendations, and phase instructions in real time.

This three-phase architecture transforms the BrainStormX assistant from a chatbot into a fully autonomous workshop conductor that tracks time, monitors engagement, suggests facilitation strategies, and writes live summaries all while the session unfolds.

Time Awareness Intelligence

We implemented a custom TimeContextProvider that enables BrainStormX to perceive and act based on time. It manages each workshop phase with precision, detects idle states (e.g., when no new ideas are being added), and prompts participants to stay engaged.

For instance, during the brainstorming phase of the workshop, if a participant ask for ideas, the AI can respond with:

“You have 5 minutes left in this phase, try adding a few more ideas before time runs out!”

It also summarizes ongoing progress and dynamically refocuses participants on the workshop’s objectives as each phase advances.

Persistent Organizational Memory

Amazon Bedrock AgentCore Memory was integrated to store every conversation, decision, and idea within an organizational knowledge graph for retrieval and contextual reasoning. This means BrainStormX no longer starts workshops from scratch, instead it take the title and objectives and then it reviews similar past sessions to frame the agenda, generate relevant guidance, and deliver facilitation tailored to each team’s style and history.

Real-Time Asynchronous Collaboration

BrainStormX is built on Flask-SocketIO and WebRTC, unifying video, audio, and text collaboration with AI-assisted facilitation. Key components include:

  • Real-time transcription using Vosk for offline speech-to-text, STT Server.
  • AI voice narration powered by Piper for natural first-person readouts, TTS Server.
  • Interactive tools (voting, chat, screen sharing, and document viewing) implemented via
  • Socket.IO Frontend, and Flask-SocketIO Backend.

Participants witness results materialize instantly, clustering, voting, and summaries appear in real time. No sticky notes. No manual boards. Just fluid, intelligent collaboration.

Technical Architecture Snapshot

  • Backend: Python 3.11, Flask, Flask-SocketIO
  • Frontend: Jinja2, Vanilla JS, WebRTC
  • AI Layer: Amazon Bedrock Nova Lite, Nova Pro, Nova Canvas
  • Memory Layer: Amazon Bedrock AgentCore Memory
  • Speech & Transcription: Vosk (offline), Piper (TTS)
  • Database: SQLite (dev) / PostgreSQL (production)
  • Deployment: EC2, Docker, Prometheus, Eventlet, Flask-Talisman

Challenges We Overcame

  • Strict JSON compliance: Nova models occasionally returned unstructured text. We implemented robust JSON extractors and Pydantic validation to maintain data integrity.

  • AgentCore SDK evolution: We adapted to shifting API schemas by dynamically adjusting runtime message formats.

  • Real-time reliability We used circuit breakers and adaptive retries to maintain a smooth experience under network fluctuations.

What we learned

  1. Context outweighs prompt engineering: embedding workshop state and time context directly into prompts improved reasoning accuracy dramatically.
  2. Tone is a facilitator’s superpower: using persona-aware responses built trust and kept conversations constructive.
  3. Telemetry ensures trust: logging Bedrock latency, retries, and model IDs made debugging transparent and efficient.

What's next for BrainStormX

  • Multi-agent orchestration, supervisor agents for sentiment, compliance, and outcome forecasting.
  • Predictive insights using AWS Forecast and QuickSight to analyze market and project feasibility directly within the workshop.
Share this project:

Updates