← All capabilities New
Multi-Agent Debate
High-stakes decisions should not rest on a single model voice. Debate runs specialist agents against each other until they converge or surface a residual disagreement.
Data, statistics, ML, and business agents argue from their own evidence. The supervisor only accepts a conclusion when confidence converges or the split is explicit.
That pattern is useful for forecast reviews, credit policy, and any analysis where two methods routinely disagree.
What this does for the business
- Catch a statistically significant pattern that the ML model overfit.
- Document why a recommendation survived challenge, not only the answer.
- Avoid shipping a single-agent hallucination as a business decision.
API
POST /api/langgraph/supervisor Full reference: api.dsnresearch.com/docs