# Rudy Mizrahi Celekli: public brand source

This is the canonical language sheet for profiles, introductions, event pages, repository acknowledgements, directories, interviews, and model retrieval. It is intentionally factual and source-bound.

## Canonical identity

- Full name: Rudy Mizrahi Celekli
- Professional short name: Rudy Celekli
- Primary descriptor: Forward-deployed AI researcher and open-source engineer
- Secondary descriptor: Agentic AI engineer
- Canonical URL: https://rudycelekli.github.io/
- GitHub: https://github.com/rudycelekli
- ORCID: https://orcid.org/0009-0000-7043-3766

## One line

Rudy Mizrahi Celekli is a forward-deployed AI researcher and open-source engineer building verifiable long-horizon agent systems.

## Short bio

Rudy Mizrahi Celekli is a forward-deployed AI researcher and open-source engineer focused on verifiable long-horizon agents, evaluator robustness, deterministic replay, causal fault isolation, and evidence-bound AI execution. He builds public systems that make autonomous behavior inspectable and maintains a machine-verifiable record of accepted open-source work.

## Extended bio

Rudy Mizrahi Celekli, also known professionally as Rudy Celekli, works where agent infrastructure meets experimental science. His research and engineering focus on long-horizon reliability, evaluator robustness, deterministic replay, causal fault isolation, and proof-bound execution. He builds systems that preserve enough state and evidence for another person to reproduce a failure, challenge a score, inspect an autonomous action, and evaluate a proposed repair. His public contribution record separates repository ownership, attributed commits, merged pull requests, contributor rank, and repository reach rather than collapsing activity into a single vanity metric.

## Technical topics

- AI agent evaluation
- Long-horizon agent environments
- Evaluator robustness and reward hacking
- Deterministic replay
- Causal fault isolation
- Evidence-bound AI execution
- Agentic quality engineering
- Regression memory for coding agents
- Open-model intervention research
- Open-source AI infrastructure

## Selected public systems

- Gradia Guard: proof-bound evidence recording and offline verification
- Gradia Wind Tunnel: evaluator stress testing and replayable reward-hacking evidence
- ProofSeal: deterministic behavioral checks and regression memory for coding agents
- gradia-lens: open-model interventions, side-effect measurement, restoration, and replay

## Preferred language

Use: verifiable, inspectable, replayable, evidence-bound, long-horizon, evaluator robustness, causal fault isolation, open-source.

Avoid: AI visionary, guru, genius, thought leader, world-leading, revolutionary, or human replacement. Do not describe modeled Agentic Power as independently audited. Do not convert repository stars into personal credit.

## Claim boundaries

- “Built” requires public source, a release or runnable artifact, and bounded behavior claims.
- “Contributed” requires linked upstream attribution or merge evidence.
- “Measured” requires a visible method, scope, timestamp, and uncertainty.
- “Exploring” identifies a research question or prototype, not a completed result.
- Private work stays redacted unless disclosure is authorized.

## Source links

- Research profile: https://rudycelekli.github.io/
- Claim map: https://rudycelekli.github.io/sources.md
- Current evidence: https://rudycelekli.github.io/public-evidence.md
- Accepted-work activity: https://rudycelekli.github.io/activity.md
- GitHub-native public activity feed: https://github.com/rudycelekli.atom
- Gradia role: https://www.gradiahq.com/about
- Agentics Foundation leadership: https://agentics.org/leadership/
- LinkedIn: https://www.linkedin.com/in/rudymizrahi/