Building systems for reliable AI/API execution.
Focused on:
- retry semantics
- execution budgets
- failure classification
- routing correctness
- production receipts
- deterministic AI orchestration
Currently building:
- Pitstop — reliability infrastructure for AI/API execution
- Pitstop Truth — documented execution failure corpus
- PracticallyAI — Private objective-agnostic execution engine built around skills, validation loops, and structured orchestration
Real reliability findings and execution fixes merged into production AI systems and developer tooling.
| Repository | Contribution |
|---|---|
 genkit |
Identified missing Retry-After propagation channel across middleware boundary — fix merged across Anthropic, OpenAI, and Gemini integrations (Issue #5270 · Merged PR #5343) |
 gstack |
Improved provider-aware Retry-After handling (Merged) |
| gbrain |
Added receipt document type to native frontmatter inference (Open PR) |
 gsd-2 |
Fixed unwired Codex execution path (Merged PR) |
 pi |
Separated retry semantics for provider failures (Merged PR) |
 Cline |
Surfaced Retry-After propagation behavior (Issue) |
 Botpress |
Identified SDK vs CLI retry policy divergence (Merged PR) |
 Superset |
Surfaced agent-side 429 propagation issue (Merged PR) |
 oh-my-claudecode |
Improved provider retry behavior handling (Merged PR) |
Most AI systems fail in the layers around the model.
The recurring problems are surprisingly consistent:
- retries without provider semantics
- cooldowns enforced at the wrong scope
- collapsed error classifications
- hidden retry amplification
- budgets treated as suggestions instead of constraints
- orchestration without receipts or verification
I’m interested in making execution:
- predictable
- inspectable
- enforceable
- auditable
The goal is simple:
execute(intent, budget, policy) -> result + receipt
A growing corpus of real-world AI/API execution failures, reliability hazards, and remediation patterns.
Current focus areas:
- 429 handling
- Retry-After semantics
- provider cooldown behavior
- routing failures
- CAP vs WAIT classification
- retry amplification
- blast-radius correctness
Solo founder. Background in systematic trading infrastructure and quantitative strategy development.
Building at the intersection of execution reliability, deterministic orchestration, and AI-native systems.