SynthMind

Enterprise Autonomous Co-Thinking Runtime on Google Cloud • Multi-Agent Research Synthesis • Adaptive Cognitive Decision Intelligence

Executive Summary

Modern engineering teams, researchers, and strategy executives face severe cognitive overload when navigating multi-variable technical trade-offs, architecture decisions, and procurement evaluations. Conventional conversational AI systems operate reactively — producing unstructured prose that lacks mathematical rigor, systematic verification, and collaborative depth.

SynthMind is an enterprise-grade autonomous collaborative co-thinking partner engineered on the Google Agent Development Kit (ADK) and powered by Gemini 3.5 / 3.7 Flash with Google Cloud Firestore:

Socratic Problem Deconstruction: Actively interrogates ambiguous requirements, surfaces unstated technical constraints, and maps latent risk factors before recommending solutions.
Real-Time Token Streaming (SSE): High-throughput Server-Sent Events delivering sub-second First Token Latency (TTFT) with live agent deliberation traces.
Deterministic Decision Science: Decoupled quantitative Multi-Criteria Decision Analysis (MCDA) engine that computes weighted scores, sensitivity sweeps, and rank sorting with exact mathematical precision.
Adversarial Self-Critique: Integrated auditor agent that audits synthesis outputs for cognitive biases, unsupported premises, and risk blindspots with calibrated confidence scoring (✓ Verified, ◐ Reviewed, ⚠ Needs Review).
Dynamic Cognitive Profiling: Continuously infers user mental models across four cognitive dimensions (Analytical, Detail, Speed, Visual) to dynamically calibrate communication density.

🏆 Google Challenge & Track Compliance Matrix

Every mandatory criterion specified across all competition tracks is 100% fulfilled and verified:

Mandatory Requirement	Enterprise Implementation in SynthMind	Verifiable Code Location / Artifact
1. Gemini 3.5 or newer (Gemini API / Vertex AI)	• Primary inference powered by Gemini 3.5 Flash Lite with automated fallback to Gemini 3.5 Flash & Gemini 3.7 Flash • Real-time factual retrieval via Google Search Grounding Tool (`google_search=GoogleSearch()`) • Dual-mode runtime: Switchable between Google AI Studio and Vertex AI via `GOOGLE_GENAI_USE_VERTEXAI`	`backend/main.py` `backend/config/settings.py`
2. Google Agent Framework (Google ADK, GenAI SDK)	• Hierarchical multi-agent runtime using Google ADK (Agent Development Kit) • Structured agent definitions with strict input/output Pydantic schemas • Root Orchestrator coordinating sub-agents (`Clarifier`, `Ingester`, `Synthesizer`, `Adapter`, `Critic`) via asynchronous state machine	`backend/core/agents/` `backend/core/agents/orchestrator.py` `backend/core/agents/critic.py`
3. Google Cloud Infrastructure (Firestore, Cloud Run, Hosting)	• Google Cloud Firestore (Datastore NoSQL) for persistent session history, decision matrix state, and user profiles (`synthmind-ai-d39ed`) • Firebase Hosting (`firebase.json`, `.firebaserc`) serving edge-optimized static web assets • Google Cloud Run production container configuration (`backend/Dockerfile`, `cloudbuild.yaml`)	`backend/adapters/firestore_adapter.py` `backend/Dockerfile` `cloudbuild.yaml` `firestore.rules`
🌟 BONUS: Google AI Models (Gemma, Veo, Lyria)	• Google Gemma 2 (`gemma-4-26b-a4b-it`) for parameter-efficient on-device edge distillation (`POST /api/gemma/distill`) • Google Veo 3.1 (`veo-3.1-generate-preview`) for multi-shot cinematic video storyboarding (`POST /api/veo/storyboard`) • Google DeepMind Lyria acoustic sonic design cues embedded in storyboard production	`backend/core/agents/gemma_agent.py` `backend/core/agents/veo_studio.py`

🏗️ System Architecture & Data Flow

graph TD
    subgraph "1. Client & Presentation Layer"
        User([👤 User / Browser])
        NextApp["Next.js 16 Client Dashboard\n(Neo-Luminescence Glassmorphism UI)"]
        FirebaseHosting["Firebase Hosting\n(Google Global CDN)"]
        User -->|Interacts| NextApp
        NextApp -.->|Served by| FirebaseHosting
    end

    subgraph "2. API Gateway & Agent Hierarchy"
        FastAPI["FastAPI Gateway (Python 3.13)\n(CORS / Rate Limiting / Structured Telemetry)"]
        Orchestrator["Google ADK Root Orchestrator\n(State Machine & Dynamic Routing)"]
        
        NextApp -->|SSE Streaming / REST / JSON| FastAPI
        FastAPI -->|Lifecycle Delegate| Orchestrator
        
        subgraph "Google ADK Sub-Agents"
            Clarifier["🔍 Clarifier Agent\n(Socratic Deconstruction)"]
            Ingester["📄 Ingester Agent\n(Multi-modal Data Parser)"]
            Synthesizer["⚡ Synthesizer Agent\n(Framework Builder)"]
            Adapter["🧬 Adapter Agent\n(Cognitive Profiler)"]
        end
        
        Orchestrator --> Clarifier
        Orchestrator --> Ingester
        Orchestrator --> Synthesizer
        Orchestrator --> Adapter
    end

    subgraph "3. AI Core & Google Foundation Models"
        GeminiFlash["Gemini 3.5 / 3.7 Flash\n(Google GenAI SDK)"]
        GoogleSearch["Google Search Grounding Tool\n(Real-time Fact Retrieval)"]
        AdversarialCritic["🛡️ Adversarial Critic\n(Bias & Confidence Audit)"]
        GemmaDistiller["💎 Google Gemma 2 Distiller\n(Edge Fact Extraction)"]
        VeoStudio["🎬 Google Veo 3.1 & Lyria Studio\n(Video Storyboards & Audio Cues)"]
        
        Orchestrator <-->|Prompt & Tool Calls| GeminiFlash
        GeminiFlash <-->|Live Web Queries| GoogleSearch
        Orchestrator -->|Concurrent Audit| AdversarialCritic
        Orchestrator -->|Distillation API| GemmaDistiller
        Orchestrator -->|Storyboard API| VeoStudio
    end

    subgraph "4. Deterministic Engines & Persistence"
        MCDAEngine["📊 Deterministic MCDA Math Engine\n(Weights, Normalized Sums, Sensitivity)"]
        FirestoreDB[("Google Cloud Firestore NoSQL Database\n(Sessions, Artifacts, User Profiles)")]
        
        FastAPI -->|Pure Python Math| MCDAEngine
        FastAPI <-->|MemoryPort / Google Cloud SDK| FirestoreDB
    end

Clean Hexagonal Architecture (Domain Core Isolation)

synthmind/
├── backend/
│   ├── adapters/          # Pluggable Storage Adapters (Google Cloud Firestore, InMemory)
│   ├── config/            # Centralized Settings, Whitelisted CORS, Environment Variables
│   ├── core/              # Domain Core (Pure Python, Zero Vendor Lock-in)
│   │   ├── agents/        # 6 Autonomous Agents (Orchestrator, Clarifier, Ingester, Synthesizer, Adapter, Critic)
│   │   ├── events/        # In-Memory Typed Event Bus for Decoupled Observability
│   │   ├── interfaces/    # Port Abstractions (MemoryPort Protocol)
│   │   ├── models/        # Pure Domain Entities (Session, Synthesis, UserProfile, DecisionMatrix)
│   │   └── tools/         # Deterministic MCDA Scoring & Sensitivity Analysis Engine
│   ├── observability/     # Structured JSON Logging & Distributed Trace Correlation
│   ├── tests/             # Automated Pytest Suite (Unit, Integration, Tools, API)
│   └── main.py            # High-Performance FastAPI Gateway & SSE Stream Pipeline
├── frontend/              # Next.js 16 Neo-Luminescence Glassmorphism UI (DOMPurify Sanitized)
│   ├── src/app/           # Next.js App Router (Streaming Chat, Sliders, Matrix View, Modals)
│   └── public/            # Static Web Assets
├── .firebaserc            # Firebase Project Identifier (synthmind-ai-d39ed)
├── firebase.json          # Firebase Hosting & Firestore Rules Configuration
├── firestore.rules        # Google Cloud Firestore Database Security Policies
├── firestore.indexes.json # Firestore Collection Group Indexes
└── README.md              # Project Documentation & Verification Guide

🤖 Multi-Agent Cognitive Hierarchy

Agent Role	Framework Implementation	Core Architectural Responsibility	Cognitive Specialization
🎯 Orchestrator	Google ADK Root Agent	Manages state machine lifecycle & dynamic routing	Workflow governance, phase transitions (`Discovery` → `Clarification` → `Ingestion` → `Synthesis` → `Feedback`)
🔍 Clarifier	Google ADK Sub-Agent	Socratic problem deconstruction & edge case probing	Constraint mapping, risk surfacing, unstated premise interrogation
📄 Ingester	Google ADK Sub-Agent	Multimodal document parsing (PDFs, raw text, web URLs)	Key fact extraction, entity recognition, evidence data point mapping
⚡ Synthesizer	Google ADK Sub-Agent	Builds structured decision matrices & frameworks	Multi-criteria scoring, SWOT grids, competitive trade-off matrices
🧬 Adapter	Google ADK Sub-Agent	Continuously evaluates user communication patterns	Cognitive radar profiling across 4 dimensions (Analytical, Detail, Speed, Visual)
🛡️ Critic	Adversarial Sub-Agent	Independent verification & cognitive bias detection	Non-blocking second-pass audit, confirmation bias detection, assumption validation

🛠️ Technology Stack

Layer / Category	Technology	Version / Model	Architectural Role & Implementation Detail
🧠 Foundation Models	Gemini Flash Family	`gemini-3.5-flash-lite`, `gemini-3.5-flash`, `gemini-3.7-flash`	High-speed multimodal inference, complex structured reasoning, cascading failover
🔍 Search Grounding	Google Search Tool	`GoogleSearch()`	Real-time web retrieval for dynamic fact-checking and temporal grounding
🤖 Agent Framework	Google ADK + GenAI SDK	`google-genai>=1.0.0`	Hierarchical agent orchestration, sub-agent delegation, structured Pydantic schemas
💎 Edge & Open Models	Google Gemma 2	`gemma-4-26b-a4b-it` / `gemma-4-31b-it`	Parameter-efficient distillation, edge fact extraction, local deployment capability
🎬 Multimodal Studio	Google Veo 3.1 & Lyria	`veo-3.1-generate-preview`	Multi-shot cinematic video storyboard generator with acoustic soundscape prompts
⚡ Backend API Gateway	FastAPI + Uvicorn	Python 3.13 / `fastapi>=0.115`	Async REST gateway, Server-Sent Events (SSE) streaming, sliding-window rate limiter
📊 Decision Engine	MCDA Python Engine	Deterministic Math Core	Weighted-sum scoring, normalized rank sorting, criterion sensitivity sweeps
🗄️ Database & Storage	Google Cloud Firestore	NoSQL Datastore Mode	Persistent multi-user session state, decision artifacts, and cognitive profiles
🎨 Frontend Web UI	Next.js + React + TS	Next.js 16 / React 19 / TypeScript 5	Dark glassmorphic dashboard, live interactive weight sliders, real-time SSE consumer
🛡️ Security & Sanitation	DOMPurify + CORS Middleware	`dompurify>=3.2.4`	Client-side HTML sanitization preventing XSS, strict origin CORS validation
🧪 Testing Framework	Pytest + Pytest-Asyncio	`pytest>=8.0.0`	100% passing test suite across tools, state machines, persistence, and API routes
🚀 Cloud Deployment	Firebase Hosting	Google Global Edge CDN	Global edge-cached static distribution with automated cache invalidation

🧪 Automated Testing Suite (100% Pass)

SynthMind includes an enterprise-grade automated test suite ensuring complete reproducibility, mathematical correctness, and system stability:

cd synthmind/backend
pytest tests/ -v

============================= test session starts =============================
platform win32 -- Python 3.13.7, pytest-9.1.1, pluggy-1.6.0
rootdir: synthmind/backend
collected 17 items

tests/test_api.py::test_health_check PASSED                              [  5%]
tests/test_api.py::test_recalculate_endpoint PASSED                      [ 11%]
tests/test_api.py::test_export_endpoint PASSED                           [ 17%]
tests/test_api.py::test_events_endpoint PASSED                           [ 23%]
tests/test_phase_transitions.py::test_initial_onboarding_transition PASSED [ 29%]
tests/test_phase_transitions.py::test_clarification_to_ingestion_transition PASSED [ 35%]
tests/test_phase_transitions.py::test_ingestion_to_synthesis_transition PASSED [ 41%]
tests/test_phase_transitions.py::test_synthesis_to_feedback_transition PASSED [ 47%]
tests/test_phase_transitions.py::test_classify_message_type PASSED       [ 52%]
tests/test_sessions.py::test_session_lifecycle_and_serialization PASSED  [ 58%]
tests/test_sessions.py::test_in_memory_adapter PASSED                    [ 64%]
tests/test_sessions.py::test_user_profile_persistence PASSED             [ 70%]
tests/test_tools.py::test_recalculate_matrix_scoring PASSED              [ 76%]
tests/test_tools.py::test_recalculate_matrix_zero_weights PASSED         [ 82%]
tests/test_tools.py::test_sensitivity_analysis_scenarios PASSED          [ 88%]
tests/test_tools.py::test_sensitivity_analysis_invalid_criterion PASSED  [ 94%]
tests/test_tools.py::test_compute_confidence_bucket PASSED               [100%]

======================= 17 passed in 3.55s ========================

🚀 Spin-up & Reproducibility Instructions

Follow this step-by-step guide to run SynthMind locally or verify the live cloud deployment.

Option A: Live Production Cloud Access (Zero Setup)

Live Web Application: https://synthmind-ai-d39ed.web.app
Cloud Infrastructure: Google Cloud Firestore NoSQL Database (synthmind-ai-d39ed)
Interactive Swagger Docs: http://localhost:8000/docs (when running backend)

Option B: Local Spin-up (Step-by-Step)

Prerequisites

Python 3.10+ (Tested on Python 3.11, 3.12, 3.13)
Node.js 18+ (Tested on Node 20 LTS, 22)
Gemini API Key from Google AI Studio

Step 1: Clone Repository

git clone https://github.com/Anurag-tech22/synthmind.git
cd synthmind

Step 2: Configure Environment Variables

# Configure Backend
cd backend
cp .env.example .env

Edit backend/.env with your credentials:

GEMINI_API_KEY=your_gemini_api_key_here
GEMINI_MODEL=gemini-3.5-flash-lite
ENABLE_FIRESTORE=false        # Set true if using Google Cloud Firestore service account
ENABLE_SEARCH_GROUNDING=true  # Enables live Google Search fact grounding

Step 3: Initialize and Run Backend

# In synthmind/backend:
python -m venv venv

# Windows PowerShell:
.\venv\Scripts\activate
# macOS / Linux:
# source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Run test suite to verify system integrity (17/17 tests must pass)
pytest tests/ -v

# Start FastAPI API Gateway
python -m uvicorn main:app --host 0.0.0.0 --port 8000

Backend is operational at http://localhost:8000 (API documentation at http://localhost:8000/docs)

Step 4: Initialize and Run Frontend Dashboard

# Open a new terminal window:
cd synthmind/frontend

# Install dependencies
npm install

# Start Next.js development server
npm run dev

Frontend is operational at http://localhost:3000

Option C: Cloud Deployment (Firebase & Google Cloud)

To re-deploy the static Next.js frontend and Google Cloud Firestore security rules to your own Firebase project:

# In project root (synthmind):
npx next build --prefix frontend
npx firebase login
npx firebase deploy

🔒 Enterprise Security Posture

XSS Immunity via DOMPurify: All LLM streaming chunks, synthesized decision frameworks, and Markdown HTML rendering pipelines pass through client-side DOMPurify.sanitize() prior to DOM injection.
Sliding-Window Rate Limiting: Backend incorporates an in-memory sliding-window rate limiter (60 requests/minute per client IP) to protect against denial-of-service and API quota exhaustion.
Strict Origin CORS Whitelisting: Cross-Origin Resource Sharing (CORS) is strictly restricted to configured development and production domains (settings.cors_origins).
Zero Secret Leakage: Private keys, Firebase service account credentials, and local environment files are strictly blocked by .gitignore.

📝 License

SynthMind is released under the MIT License.

Navigation Menu

Repository files navigation

SynthMind

Executive Summary

🏆 Google Challenge & Track Compliance Matrix

🏗️ System Architecture & Data Flow

Clean Hexagonal Architecture (Domain Core Isolation)

🤖 Multi-Agent Cognitive Hierarchy

🛠️ Technology Stack

🧪 Automated Testing Suite (100% Pass)

🚀 Spin-up & Reproducibility Instructions

Option A: Live Production Cloud Access (Zero Setup)

Option B: Local Spin-up (Step-by-Step)

Prerequisites

Step 1: Clone Repository

Step 2: Configure Environment Variables

Step 3: Initialize and Run Backend

Step 4: Initialize and Run Frontend Dashboard

Option C: Cloud Deployment (Firebase & Google Cloud)

🔒 Enterprise Security Posture

📝 License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
backend		backend
frontend		frontend
.firebaserc		.firebaserc
.gitignore		.gitignore
README.md		README.md
architecture-diagram.png		architecture-diagram.png
banner.png		banner.png
cloudbuild.yaml		cloudbuild.yaml
firebase.json		firebase.json
firestore.indexes.json		firestore.indexes.json
firestore.rules		firestore.rules
setup.bat		setup.bat
test_gemini.py		test_gemini.py

Latest commit

History

Folders and files

Repository files navigation

SynthMind

Executive Summary

🏆 Google Challenge & Track Compliance Matrix

🏗️ System Architecture & Data Flow

Clean Hexagonal Architecture (Domain Core Isolation)

🤖 Multi-Agent Cognitive Hierarchy

🛠️ Technology Stack

🧪 Automated Testing Suite (100% Pass)

🚀 Spin-up & Reproducibility Instructions

Option A: Live Production Cloud Access (Zero Setup)

Option B: Local Spin-up (Step-by-Step)

Prerequisites

Step 1: Clone Repository

Step 2: Configure Environment Variables

Step 3: Initialize and Run Backend

Step 4: Initialize and Run Frontend Dashboard

Option C: Cloud Deployment (Firebase & Google Cloud)

🔒 Enterprise Security Posture

📝 License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages