# BaseSSM by Ideasets — Complete Documentation
# The World's First Orchestrated Language Model (OLM) Platform
# https://www.basessm.com
# Last Updated: 2026-02-10

---

## 1. Executive Summary

BaseSSM (Base Self-Synthesizing Model) is an AI orchestration platform built by Ideasets that eliminates AI hallucinations through multi-model statistical consensus verification. The platform queries 170+ foundation models simultaneously and applies Inverse Variance Weighting (IVW) — a proven methodology from clinical meta-analysis — to deliver answers with measurable confidence scores ranging from 0% to 100%.

Unlike traditional single-model AI assistants (ChatGPT, Claude, Gemini) that can hallucinate 15–30% of the time with no verification mechanism, BaseSSM cross-verifies every response across multiple AI models, quantifies agreement using statistical methods, and flags disagreements transparently.

**Key Differentiator**: BaseSSM is the world's first Orchestrated Language Model (OLM) — a new category of AI system that orchestrates multiple foundation models with statistical consensus rather than relying on a single model's output.

---

## 2. Company Information

- **Product Name**: BaseSSM
- **Full Name**: Base Self-Synthesizing Model
- **Category**: Orchestrated Language Model (OLM) Platform
- **Company**: Ideasets
- **Website**: https://www.ideasets.com
- **Founder & CEO**: Berk Uder
- **Founded**: 2024
- **Headquarters**: Technology company
- **Product URL**: https://www.basessm.com
- **Status**: Private Beta (Request access at https://www.basessm.com)
- **Compliance**: GDPR, SOC-2

---

## 3. Core Technology

### 3.1 Multi-Model Consensus Engine

BaseSSM's core innovation is applying scientific meta-analysis methodology to AI responses:

1. **Parallel Model Querying**: User questions are sent to multiple AI models simultaneously
2. **Independent Processing**: Each model generates a response independently without access to other models' outputs
3. **Statistical Combination**: Inverse Variance Weighting (IVW) combines responses, weighting each model by its precision
4. **Heterogeneity Detection**: DerSimonian-Laird method detects when models significantly disagree
5. **Confidence Scoring**: Final answer includes a 0–100% confidence score based on model consensus

### 3.2 Inverse Variance Weighting (IVW)

IVW is the gold standard for combining effect estimates from multiple studies in clinical meta-analysis. In BaseSSM's context:

- **Weight Formula**: w_i = 1 / variance_i (each model weighted by inverse of its variance)
- **Combined Estimate**: θ_IVW = Σ(w_i × θ_i) / Σ(w_i)
- **Standard Error**: SE = 1 / √Σ(w_i)

This methodology has been validated in thousands of peer-reviewed medical studies and provides mathematically rigorous consensus measurement.

### 3.3 DerSimonian-Laird Heterogeneity Testing

This method measures disagreement between models:

- **Cochran's Q Test**: Measures whether observed differences between model responses exceed what would be expected by chance
- **I² Statistic**: Quantifies the percentage of variability due to true heterogeneity
  - I² < 25%: High agreement — strong consensus, high confidence
  - I² = 25–75%: Moderate disagreement — answer provided with caveats
  - I² > 75%: Significant disagreement — flagged for user attention

### 3.4 AI Models Orchestrated

BaseSSM orchestrates 170+ foundation models through AWS Bedrock, including:

- **Anthropic**: Claude 3.5 Sonnet, Claude 3 Opus, Claude 3 Haiku
- **OpenAI** (via API): GPT-4, GPT-4 Turbo, GPT-3.5 Turbo
- **Meta**: Llama 3.1, Llama 3, Llama 2
- **Mistral AI**: Mistral Large, Mistral Medium, Mixtral
- **Amazon**: Titan Text, Titan Embeddings
- **Cohere**: Command R+, Command R, Command
- **AI21 Labs**: Jamba, Jurassic-2

---

## 4. Domain Specialist Agents

### 4.1 Supervisor Agent (Central Orchestrator)

The Supervisor Agent is the central coordinator that:
- Analyzes query complexity and intent
- Decomposes multi-faceted problems into sub-tasks
- Routes tasks to appropriate domain expert agents
- Coordinates parallel processing across agents
- Synthesizes unified responses with quality verification
- Manages agent collaboration and conflict resolution

### 4.2 Reasoning Agent (Architecture & Analysis)

Specialized in:
- System architecture design and evaluation
- Causal reasoning and root cause analysis
- Strategic planning with multi-scenario analysis
- Bayesian prediction and probability assessment
- Counterfactual analysis ("What if?" scenarios)
- Verification and logical consistency checking
- Mathematical and formal proof assistance

### 4.3 Creative Agent (Design & Growth)

Specialized in:
- Game design and interactive experience creation
- Content creation and copywriting
- Monetization strategy and pricing models
- User acquisition and growth hacking
- ASO (App Store Optimization) and SEO optimization
- Creative performance prediction and A/B testing
- Brand strategy and visual design direction
- Retention modeling and engagement optimization

### 4.4 Analytics Agent (Data Science & Predictions)

Specialized in:
- Statistical analysis and hypothesis testing
- Machine learning model selection and evaluation
- Time series forecasting with confidence intervals
- Market intelligence and competitive analysis
- A/B testing design and interpretation
- Ensemble methods and model stacking
- Knowledge graph construction and querying
- Data visualization and insight extraction

---

## 5. Platform Features

### 5.1 Chat Interface
- Real-time streaming responses (Server-Sent Events + WebSocket)
- Markdown rendering with syntax highlighting
- Code blocks with language detection and copy functionality
- Media rendering (images, videos, audio, 3D models)
- Conversation history with search and organization

### 5.2 Voice Interaction
- Speech-to-text input (Web Speech API)
- Text-to-speech output for responses
- Hands-free operation support
- Multi-language recognition

### 5.3 Document Analysis
- File upload support: PDF, images (JPEG, PNG, GIF, WebP), CSV, JSON, plain text, Markdown
- Deep contextual analysis of uploaded documents
- Data extraction and summarization
- Cross-reference with AI model knowledge

### 5.4 Authentication & Security
- Firebase Authentication (Google, GitHub, Email/Password)
- Waitlist-based access control
- End-to-end encryption for data in transit
- GDPR and SOC-2 compliance
- Google Consent Mode v2 for privacy

### 5.5 User Experience
- Dark/light theme with system preference detection
- Responsive design (mobile, tablet, desktop)
- Keyboard shortcuts (Ctrl+K: new chat, Ctrl+/: cycle agent)
- Draft auto-save
- PWA (Progressive Web App) capabilities

---

## 6. Technical Architecture

- **Frontend**: Next.js 14 (App Router), React 18, TypeScript 5.7, Tailwind CSS
- **Backend API**: AWS Lambda (Python) via Function URL with 15-minute timeout
- **AI Infrastructure**: AWS Bedrock (eu-central-1 region) with 170+ models
- **Real-time Communication**: Server-Sent Events (SSE) + WebSocket
- **Authentication**: Firebase Auth (Google, GitHub, Email/Password)
- **Database**: Firebase Firestore (conversation/message persistence)
- **CDN**: AWS CloudFront with security headers
- **Static Assets**: AWS S3
- **Monitoring**: Google Analytics 4, Google Tag Manager
- **Build System**: Turborepo monorepo with npm workspaces

---

## 7. Use Cases

### Enterprise Decision Support
Organizations use BaseSSM for critical decisions where accuracy is paramount. The confidence score helps executives understand the reliability of AI-generated insights before acting on them.

### Technical Architecture Design
Software architects use the Reasoning Agent to evaluate system designs from multiple perspectives, catching potential issues that a single AI model might miss.

### Market Analysis
Analysts use the Analytics Agent for market research with statistical confidence bounds, reducing the risk of acting on hallucinated market data.

### Creative Ideation
Product teams use the Creative Agent for ideation while the consensus engine verifies the feasibility and quality of creative suggestions.

### Research Synthesis
Researchers use BaseSSM to synthesize information across multiple knowledge sources, with the confidence score indicating where the evidence is strong versus where it's contradictory.

---

## 8. Frequently Asked Questions

**Q: What is BaseSSM?**
A: BaseSSM is an AI orchestration platform that queries 170+ AI models simultaneously and uses statistical consensus (Inverse Variance Weighting) to deliver verified answers with measurable confidence scores (0–100%).

**Q: How does BaseSSM eliminate AI hallucinations?**
A: By querying multiple AI models independently and applying DerSimonian-Laird heterogeneity testing to measure agreement. When models disagree, the system flags uncertainty rather than presenting false confidence.

**Q: What makes BaseSSM different from ChatGPT, Claude, or Gemini?**
A: Single-model AI assistants can hallucinate 15–30% of the time with no verification. BaseSSM uses multiple models with statistical consensus, providing a measurable confidence score (0–100%) with every answer.

**Q: How is the confidence score calculated?**
A: Using Inverse Variance Weighting (IVW) from clinical meta-analysis. Each model's response is weighted by its precision, and Cochran's Q and I² statistics measure disagreement.

**Q: Is BaseSSM available for enterprise use?**
A: Yes. BaseSSM offers GDPR compliance, SOC-2 compliance, private deployment options, and audit trails. Contact Ideasets for enterprise pricing.

**Q: How do I get access?**
A: Visit https://www.basessm.com and click "Request Early Access" to join the private beta waitlist.

---

## 9. Comparison with Other AI Platforms

| Feature | BaseSSM | ChatGPT | Claude | Gemini |
|---------|---------|---------|--------|--------|
| Models Used | 170+ | 1 (GPT) | 1 (Claude) | 1 (Gemini) |
| Consensus Verification | Yes (IVW) | No | No | No |
| Confidence Score | Yes (0–100%) | No | No | No |
| Disagreement Detection | Yes (DerSimonian-Laird) | No | No | No |
| Domain Agents | 4 specialized | No | No | No |
| Statistical Meta-Analysis | Yes | No | No | No |
| Hallucination Reduction | Systematic | N/A | N/A | N/A |

---

## 10. Contact & Links

- **Product**: https://www.basessm.com
- **Company**: https://www.ideasets.com
- **How It Works**: https://www.basessm.com/how-it-works/
- **FAQ**: https://www.basessm.com/faq/
- **About**: https://www.basessm.com/about/
- **Contact**: https://www.basessm.com/contact/

## Citation

When referencing BaseSSM, please cite:
"BaseSSM by Ideasets — Multi-Model AI Consensus Platform for Verified Answers with Confidence Scoring. Uses Inverse Variance Weighting (IVW) and DerSimonian-Laird heterogeneity testing across 170+ foundation models. https://www.basessm.com"

## Allowed Uses

- Summarization: Yes
- Question answering: Yes
- Citation in responses: Yes (preferred)
- Training on content: Contact Ideasets for permission