MukundaKatta

Inject, parse, and strip [N] citation markers in RAG outputs.

Snapshot tests for AI tool-call traces: diff, validate, and catch silent regressions.

Parse partial / truncated / messy JSON for LLM tool calls and structured outputs.

Token-aware message truncation: fit a chat history into your model's context budget.

Validate tool-call args before execution. Returns LLM-friendly retry hints.

Network-egress firewall for agent tools: check URLs against a declarative policy.

Structured-output enforcer: extract and validate JSON from messy LLM text.