§ Research
Publications · Directions
Papers written like buildings — foundation first, ornament last.
Research at Arkashri exists to answer a single question: what would this system look like if we designed it as if it had to survive a court, a regulator and a century?
Ongoing research directions
Multi-agent audit reasoning
How to make an ensemble of specialized agents produce a single, defensible audit finding.
Cryptographic provenance for AI outputs
Signing the entire inference path — model, prompt, retrieval — as one evidence bundle.
Sovereign compliance graphs
Compiling Indian regulatory text into machine-verifiable obligations.
Deterministic replay for LLM workflows
Bit-exact rehydration of agent runs, for regulator and courtroom use.