Skip to content
View solryu91's full-sized avatar

Block or report solryu91

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
solryu91/README.md

Hi, I'm Sol

I build local-first AI systems, evaluation harnesses, privacy-conscious tools, and human-in-the-loop workflows. My work focuses on making AI behavior inspectable: explicit permissions, provenance, reproducible evidence, careful failure handling, and clear boundaries between a prototype and a supported claim.

Selected work

Kabbalistic Core

An inspectable cognitive-graph research prototype with deterministic replay, synthetic retrieval, bounded local-model integration, append-only continuity and correction slices, and traceable outputs. Python and TypeScript/React; 145 tests in the public edition.

Agent Security Lab

A standard-library Python harness for indirect prompt-injection research in tool-using agents. It compares a deliberately vulnerable dispatcher with exact- argument permission enforcement, records ordered evidence, builds replayable run bundles, and separates attempted, blocked, hallucinated, inconclusive, and corroborated simulated outcomes. 59 tests.

Chat Export Toolkit

Dependency-free, privacy-first utilities for inspecting a ChatGPT data export locally. The public repository uses a fictional fixture and keeps generated indexes and real exports outside the clone. 4 end-to-end tests.

Windows Security Posture

A read-only PowerShell posture collector with documentation, defensive ignore rules, CI, and a static safety check. It measures selected Windows security settings without applying remediation.

AI systems case studies

Public-safe architecture notes covering a provenance-aware Shared Memory Spine, a 37-track/53-idea project atlas, a privacy-conscious full-stack service website, and an AI-assisted creator workflow with media QA and publication gates.

How I work

  • I use synthetic fixtures and local-first defaults when real data is not needed.
  • I separate observed evidence from inference and historical status from current state.
  • I design consequential writes, durable memory, and external actions around explicit human authority.
  • I keep failures, limitations, and unverified boundaries visible instead of polishing them out of the story.

Technologies

Python · TypeScript · React · Next.js · PowerShell · GitHub Actions · knowledge graphs · retrieval and provenance systems · AI evaluation · agent security · privacy-by-design · technical documentation

I built these projects through AI-assisted engineering workflows while retaining human ownership of product direction, review, and release decisions.

Pinned Loading

  1. kabbalistic-core kabbalistic-core Public

    Inspectable local-first AI architecture with cognitive graphs, deterministic replay, synthetic retrieval, continuity experiments, and traceable evaluation.

    Python

  2. agent-security-lab agent-security-lab Public

    Synthetic indirect prompt-injection lab with exact permissions, ordered evidence, replayable runs, and 59 tests.

    Python

  3. chat-export-toolkit chat-export-toolkit Public

    Privacy-first, dependency-free tools for inspecting a ChatGPT data export locally using fictional public fixtures.

    Python

  4. ai-systems-case-studies ai-systems-case-studies Public

    Public-safe case studies in AI memory, provenance, knowledge graphs, product delivery, and human review.

  5. windows-security-posture windows-security-posture Public

    Read-only PowerShell snapshot of selected Windows security settings with a dependency-free static safety check.

    PowerShell