# Rudy Mizrahi Celekli

Rudy Mizrahi Celekli, also known professionally as Rudy Celekli, is a forward-deployed AI researcher and open-source engineer. His work focuses on verifiable long-horizon agent environments, evaluator robustness, deterministic replay, causal fault isolation, evidence-bound execution, agentic quality engineering, and open-source AI infrastructure.

Canonical profile: https://rudycelekli.github.io/

## Research question

How do we know an agent completed the right work, under the right world state, for reasons that survive independent review?

## Technical program

1. Long-horizon reliability: preserving intent, state, and recoverability across interruptions, tool failures, and changing environments.
2. Evaluator robustness: separating genuine task improvement from reward hacking, proxy optimization, judge drift, and incomplete measurement.
3. Causal fault isolation: reducing a failure to the smallest reproducible causal slice and replaying it against proposed repairs.
4. Proof-bound execution: retaining enough evidence for another person to audit an autonomous action without trusting the agent's narration.

## Selected public systems

- Gradia Guard: https://github.com/rudycelekli/gradia-guard
- Gradia Wind Tunnel: https://github.com/rudycelekli/gradia-wind-tunnel
- ProofSeal: https://github.com/rudycelekli/proofseal
- gradia-lens: https://github.com/rudycelekli/gradia-lens

## Public evidence

The public contribution ledger separates repository ownership, attributed commits, merged pull requests, accepted line changes, and GitHub-listed contributor status. It is refreshed automatically. Repository stars describe a repository's reach, not Rudy's personal credit.

- Current evidence in Markdown: https://rudycelekli.github.io/public-evidence.md
- Current evidence in JSON: https://rudycelekli.github.io/data/public-evidence.json
- Primary GitHub profile: https://github.com/rudycelekli
- Claim and source map: https://rudycelekli.github.io/sources.md

## Claim policy

A passing test is not deployment. A local demonstration is not external acceptance. Repository stars are not authorship. An agent report is not proof of completion. Private work is represented only as a redacted aggregate unless disclosure is authorized. Public ownership and upstream contribution remain separate, and uncertainty remains visible.

## Persistent identities

- GitHub: https://github.com/rudycelekli
- LinkedIn: https://www.linkedin.com/in/rudymizrahi/
- ORCID: https://orcid.org/0009-0000-7043-3766
- Gradia role: https://www.gradiahq.com/about
- Agentics Foundation leadership: https://agentics.org/leadership/