# Augustin Chan
> Independent researcher on when an AI agent can safely act alone. Writing about AI architecture, agent safety, and software engineering.

## Essays
- [Teaching Taste to an Agent](https://augustinchan.dev/posts/2026-02-14-teaching-taste-to-agents-yilin-image-pipeline): Encoding art direction as reusable agent skills for 4,096 AI-generated ink paintings from Han dynasty poetry
- [The Junzi Hypothesis](https://augustinchan.dev/posts/2026-01-25-junzi-alignment-initial-weights-hypothesis): Whether initial model weights could predispose toward naturally aligned behavior using Confucian virtue ethics
- [Seeded I-Ching Engine](https://augustinchan.dev/posts/2025-12-01-seeded-iching-engine): Open-source deterministic hexagram generation for verifiable compute using Mulberry32 PRNG
- [Memory Systems and the Graph That Wasn't](https://augustinchan.dev/posts/2025-10-28-memory-systems-and-the-graph-that-wasnt): JSONB conversations vs graph databases for AI agent memory systems
- [LLM Reasoning Patterns and Failure Modes](https://augustinchan.dev/posts/2025-09-27-llm-reasoning-pattern-classification-failure-modes): Meta-classification failure modes where LLM reasoning converges to templates despite varied chain-of-thought
- [DSPy, Voice, and the Evolution of Authenticity](https://augustinchan.dev/posts/2025-09-02-dspy-voice-evolution-authenticity): Multi-dimensional voice lattices with Pareto optimization for AI content generation

## Technical Posts
- [Vibe-Coded Evals: LLM as Judge](https://augustinchan.dev/posts/2025-12-13-vibe-coded-evals-llm-as-judge): Using LLMs to evaluate LLM output quality with structured rubrics
- [Whisper Cantonese Transcription](https://augustinchan.dev/posts/2026-01-13-whisper-cantonese-transcription-zh-not-yue): Force zh not yue — counter-intuitive finding for Cantonese speech-to-text
- [DSPy OpenRouter Integration](https://augustinchan.dev/posts/2025-07-08-dspy-openrouter-integration): Custom headers for failover and app attribution in DSPy with OpenRouter
- [Proust Multi-Model Comparison](https://augustinchan.dev/posts/2025-11-18-proust-multimodel-comparison): Testing 12 AI image models through fal.ai for artistic fidelity

## Projects
- [8-Bit Oracle](https://8bitoracle.ai): Tech-noir I-Ching divination app with AI interpretation, voice input, Web3 wallet integration, and Polymarket event consultation
- [QDayAnon](https://qdayanon.com): AI research platform for quantum computing with DSPy voice evolution and 170+ curated papers
- [seeded-iching-engine](https://github.com/augchan42/seeded-iching-engine): Open-source deterministic PRNG hexagram generation library
- [Plum Blossom Computer](https://github.com/augchan42/pbc-consensus-hk-2026): Consensus HK 2026 hackathon project

## Blog Archive
- [Full archive](https://augustinchan.dev/blog): 54 posts spanning 2023-2026 covering AI, Web3, React, and software architecture

## About
- Role: Independent researcher
- Focus: When can an AI agent safely act alone — risk assessment, multi-agent safety research (arXiv), agentic risk frameworks
- Background: Enterprise architecture at Informatica and Dun & Bradstreet, leading Fortune 500 customers in a global role with an APAC focus
- Education: BS Cognitive Science with Specialization in Computation, UC San Diego
- Contact: aug@digitalrain.studio
- GitHub: https://github.com/augchan42
- X/Twitter: https://x.com/augchan42