I build and explore open-source AI and machine learning tools, with a focus on dependable agent systems, model evaluation, retrieval, and AI security.
- AI agents and reasoning: tool use, planning, memory, controlled workflows, and evaluation.
- Machine learning: data quality, reproducible experiments, calibration, drift, and useful baselines.
- LLMs and RAG: retrieval quality, grounded answers, citations, and provider-independent applications.
- AI safety and security: permission boundaries, adversarial testing, auditability, and human review.
- AI infrastructure and MLOps: deployable services, observability, model lifecycle, and cost measurement.
My research interests include artificial superintelligence (ASI) and the questions it raises about reasoning, alignment, evaluation, and safe autonomy. This is a research direction, not a claim that these projects implement superintelligence.
Fixes and improvements I’ve submitted to open-source projects.
- lyogavin/airllm — Load single-file checkpoints once upfront during splitting and handle non-numeric shard names (#380)
- KittenML/KittenTTS — Split long ONNX inputs to respect the 512-token BERT limit (#162)
- JuliusBrussee/caveman — fix(antigravity-cli): declare contextFileName in agy plugin.json (#1289)
- MadsLorentzen/ai-job-search — fix(job-scraper): separate Portal CLI search terms from WebSearch site: fallback (#536) (#600)
- rohitg00/ai-engineering-from-scratch — Add Summary cheat-sheet section support to lesson pages (#622)
- affaan-m/ECC — fix: respect CLAUDE_CONFIG_DIR with WSL Windows-path conversion (#3602)
- ShawnPana/phone-harness — docs: add CHANGELOG.md and pin published version in install.md (#144)
- tashfeenahmed/freellmapi — fix(model-discovery): route custom-endpoint models with modality metadata to their native table (#1506) (#1556)
- SegFault42/HeliosGen — Allow tempfile.aiquickdraw.com in /api/download origin allow-list (issue #30) (#40)
- morluto/rea — fix(hopper): diagnose MCP provider failures with stage, state, and request context (#1044)
More contributions
- PurpleDoubleD/locally-uncensored — Preserve fenced code in compact conversation summaries (#154)
- addyosmani/agent-skills — fix(linter): ignore comments and nested metadata keys in skill frontmatter (#658)
- JayWebtech/autoshorts — fix: escape drive-letter colons in caption font paths (#48)
- ifixai-ai/iFixAi — fix(cli): allow validating built-in fixture names in validate command (#227)
- 666ghj/MiroFish — fix: order simulation history before applying the result limit (#856)
- cloudflare/security-audit-skill — Allow multiple local checks in an archived attempt to share an artifact (#70)
- Panniantong/Agent-Reach — fix(xueqiu): validate limit and clamp max size in get_hot_stocks and search_stock (#795)
- chrisvo/shipcari — Upgrade OS version (#2)
- prism lab alan vo — Work in progress: prism-lab-alan-vo
- tensor foundry alan vo — Work in progress: tensor-foundry-alan-vo
- orbit engine alan vo — Work in progress: orbit-engine-alan-vo
- vector forge alan vo — Work in progress: vector-forge-alan-vo
- cobalt hub alan vo — Work in progress: cobalt-hub-alan-vo
- axiom lens alan vo — Work in progress: axiom-lens-alan-vo
- axiom workbench alan vo — Diffusion model workflow orchestration platform with model registry, graph editor, and local inference engine
- security posture scorer alan vo — Work in progress: security-posture-scorer-alan-vo
I use AI-assisted development. Useful evidence is in the implementation: working workflows, architecture decisions, reproducible tests, measured results, and explicit limitations. Repository READMEs should make these easy to inspect; repository count alone is not a measure of quality.
I am interested in AI/ML engineering, applied AI, and AI platform opportunities.
Alan Vo · alanvo@gmail.com · GitHub