I build AI systems that keep goals, evidence, permissions, and acceptance criteria visible. My work connects research questions to tools, memory, and products people can actually use.
I care about what a model can do. I care even more about why we should trust the result.
|
A private, evidence-grounded personal journal. Photos, words, moods, time, and stories live on a freeform notebook while every AI-organized memory remains traceable to its source. Current boundary: Public, sanitized web experience backed by a privately deployed memory service. Open the live notebook ↗ · Private deployment ↗ · Public source ↗ |
|
| System | Why it exists | Route |
|---|---|---|
| ScholarFlow | A local-first research planning assistant that turns an ambiguous topic into evidence, experiments, claims, and testable stage contracts. | Open ↗ · Public source ↗ |
| Aiznoyer Atlas | A searchable knowledge garden connecting years of notes, research writing, algorithms, and project records. | Open ↗ · Public source ↗ |
| Let Code in Pad | An offline-first knowledge tool for deep desktop authoring and lightweight tablet review. | Open ↗ |
| Reliable AI Retrieved evidence, semantic geometry, and feedback-driven checks for low-confidence outputs. |
Auditable Agents Goals, permissions, evidence, and acceptance criteria expressed as workflow contracts. |
Knowledge Infrastructure Notes, experiments, projects, and decisions organized into systems that can keep growing. |
The Atlas connects long-form writing, algorithm practice, and a knowledge graph extracted from more than 100 documents.
Enter the Atlas ↗ · Read the writing ↗
🧪 Open my lab — research questions, public work, and live metrics
- How can an autonomous agent make permissions, evidence, and acceptance boundaries explicit?
- How can semantic space and source relationships expose RAG errors that merely look correct?
- As code becomes abundant, how should human advantage shift toward framing, constraints, and verification?
| Upstream | Contribution |
|---|---|
| NevaMind-AI/memUBot | Added Ollama and OpenAI-compatible local-model support; PR #6 merged. |
| NevaMind-AI/memU | Submitted OneBot v11 / NapCat QQ integration and storage fixes; PR #369 under review. |
| MonsterNone/tmall-miao | Implemented an Android uiautomator2 version and wrote the Android operation guide. |
🎮 Play with my GitHub — memory shard, alter ego, and contribution snake
A useful memory should preserve the path back to what actually happened.
Chronicle Memory design note
- Anime and Arknights live in the lab as easter eggs, not credentials.
- Guitar practice keeps the system builder human.
- Public memory shards are hand-selected; private Chronicle data is never queried here.
🛠️ Remix this profile — config, modules, and safe defaults
This profile is generated from profile.config.yml. Edit the config, run npm run generate, and the bilingual pages and SVG assets update together.
profile.config.yml: identity, projects, palette, links, and module switchescontent/public-memories.yml: hand-approved public memory shardsscripts/generate-profile.mjs: deterministic bilingual Markdown and SVG generator.github/workflows/: profile build, metrics snapshot, and contribution snake
Built as a small, forkable profile engine. Public data may update automatically; private Chronicle data never enters this repository.