Chorus replaces expensive human analyst teams with an ensemble of 17 AI specialists, each approaching your question from a different domain. The result is a calibrated probability estimate with transparent reasoning — not a black box, not a single model's opinion.
Key Features
- >17-Specialist Ensemble: Reasoner, Intelligence Analyst, Market Analyst, Researcher, Data Scientist, and more — each with a domain-specific analytical playbook
- >Parallel Fan-Out: All specialists run simultaneously via asyncio — full ensemble in the time it takes one specialist
- >TOML-Driven Playbooks: Each specialist's analytical framework is loaded from a manifest at runtime — update the playbook, update the behavior
- >Confidence-Weighted Aggregation: High-confidence specialists count more; low-confidence signals are down-weighted automatically
- >Consensus Labels: `strongly_yes` / `lean_yes` / `uncertain` / `lean_no` / `strongly_no` — human-readable in addition to the probability
- >Synthesis Pass: Optional final LLM call explains where specialists agreed, disagreed, and what the key residual uncertainty is
- >Full Reasoning Chains: Every specialist returns its step-by-step analysis, not just a number
- >REST API: Single `POST /v1/forecast` call returns the complete ensemble
How It Works
- 1.1. Submit a binary probabilistic question: *"Will the Fed cut rates before September 2026?"*Submit a binary probabilistic question: *"Will the Fed cut rates before September 2026?"*
- 2.2. Chorus fans it out to your chosen specialists in parallelChorus fans it out to your chosen specialists in parallel
- 3.3. Each specialist applies its domain playbook and returns `{probability, confidence, reasoning, assumptions, uncertainties}`Each specialist applies its domain playbook and returns `{probability, confidence, reasoning, assumptions, uncertainties}`
- 4.4. Responses aggregate: confidence-weighted average probability, consensus label, optional synthesisResponses aggregate: confidence-weighted average probability, consensus label, optional synthesis
- 5.5. You receive the ensemble result + every specialist's individual takeYou receive the ensemble result + every specialist's individual take
Use Cases
- >Hedge Funds: Calibrated probability estimates for macro and market events
- >Strategy Consultants: Quantify scenario likelihood for client presentations
- >Political Risk Analysts: Multi-perspective assessment of geopolitical outcomes
- >Researchers: Structured elicitation of AI domain knowledge across specialties
Revival Story
Cindicator raised $15M to combine crowd-sourced human predictions with ML for institutional-grade forecasts. The crowd was too noisy and expensive. Chorus replaces the human crowd with 17 AI specialists — faster, cheaper, and more consistent.
Tech Stack
FastAPI, Python, Anthropic Claude API, asyncio parallel execution, TOML manifest system. Part of the AIOS specialist mesh.