Sanscritic is a neolab working on building recursive self improving meta reasoning for expert work. Sanscritic's metareasoner technology excels in scenarios where reasoning is context and judgment heavy, may involve cross domain reasoning, and where the reasoner needs to resolve through conflicts and contradictions. The metareasoner is a neural network on top of an open weight large language model, where in the complex reasoning is handled by the metareasoner while the LLM provides the factual knowledge and information search. 

Sanscritic doesn't use reinforcement learning and hence doesn't require extensively curated datasets or reward models. All it needs are a 3-5 questions that you would normally ask it, in order for it to build a custom reasoner for an expert task. Once it has those 3-5 questions it uses its recursively self improving metareasoning technology to build and improve both the metareasoner as well as tune the under open weight LLM for greater accuracy.

We believe that in expert work reasoning is incidental to getting the right answer. Reasoning is the product. Sanscritic through its structured multi perspective reasoner provides a much greater level of transparency, auditability and forkability to the reasoning process. 

Sanscritic

Org chart

Teams

Offices

Sanscritic

Org chart

Teams

Offices

Related companies