I am Kishan

I work on building AI systems that does discovery and accelerate science; More specifically, working on causal representation learning and mechanistic transfer. Interests include studying about time, human evolution and society in the age of super intelligence. You will find some of my work on distributed systems & computational mathematics too.

Research & papers

04
2026Paper

Root Intervention Flattening

Counterexamples and corrected conditions for quotient-graph identifiability under paired interventions in causal representation learning.

Paper
2026Formal proof

Monochromatic Quantum Graphs

Formal and computational work on the Krenn–Gu monochromatic perfect-matching problem, including an audited Lean proof of the exact complex-weight no-solution result for the six-vertex, four-color slice.

Proof & audit
2026Research note

Hybrid Semantic Cache: Cache Invalidation for High-Dimensional Similarity Search

A correctness-first approach to similarity caching, using geometric bounds and explicit freshness semantics to avoid unnecessary invalidation.

Read note
2026Paper & benchmark

KyroBench: Evaluating Context Correctness Before AI Agents Act

A benchmark for whether an agent’s context remains fresh, correctly scoped, resistant to pollution, and independently verifiable as source knowledge changes.

Paper Benchmark Code

Software Engineer, Scale AI

Worked on RLHF projects, AI-agent training and evaluation, automation workflows, and coding-task model assessment.