Founder-led security research and engineering
I work with protocol teams and builders of tool-using AI systems to examine code, authority, proof, and execution boundaries. When failures occur on-chain, I reconstruct what the available evidence can support and state what remains unresolved.
Engagements
Each engagement begins with the system, revision, authority model, exclusions, and delivery window. The output is designed to support engineering decisions, not to create a false guarantee of safety.
For smart contracts, protocol components, and proof-verifier integrations preparing for release or material change.
For AI systems that use tools, retrieve untrusted content, execute code, access sensitive data, or propose consequential actions.
For confirmed on-chain events requiring a bounded technical record separated from attribution and unsupported narrative.
Selected work
The portfolio covers completed technical reports, versioned review workflows, agent authority cases, incident-reconstruction infrastructure, proof-integration analysis, and research data engineering.
Reports
Three published protocol reviews with explicit scope, findings, and limitations.
Methods
Versioned, evidence-gated workflows for EVM, proof-verifier, and zero-knowledge review.
Agent Security
Reproducible recipient-substitution and approval-binding case for transaction authority.
Incident R&D
Read-only EVM incident reconstruction with bounded evidence collection, asset-flow records, and provenance.
ZK Security
Cross-layer analysis of verifier-to-application trust boundaries in zero-knowledge proof integrations.
Infrastructure
Reproducible collection, content deduplication, FTS5 indexing, and RAG export for local security research.
Method
Scope and evidence quality control the conclusion. Unsupported hypotheses remain observations; unavailable evidence remains a limitation.
01 / Scope
Record the revision, architecture, permissions, assumptions, exclusions, and decision the work must support.
02 / Model
Identify state, trust, proof, tool, identity, and execution boundaries before testing failure paths.
03 / Challenge
Use source-connected reasoning, adversarial cases, tests, traces, and independent challenge proportional to impact.
04 / Report
Deliver prioritized findings, reproducible evidence, remediation guidance, and explicit unresolved gaps.
Contact
For a scoped conversation, include the current stage, repository or architecture, target delivery window, and any disclosure constraints. Do not send credentials or private source through a public GitHub issue.