AI Builder

Slava Spitsyn — AI Builder

Production AI systems. 52 projects. 1 person.
Live activity below — these are my real servers.

1 email / week. No fluff. Unsubscribe anytime.

slava.live

Live

[17:04:31] claude-code · security-hook caught injection attempt

[17:04:37] OpenClaw · TG session #916 → response 0.6s

[17:04:43] getward · ward.json fetched by ClaudeBot

[17:04:49] Hermes · CRM upsert → partners/jacob.md

[17:04:55] Hermes · CRM upsert → partners/sasha.md

You're watching my production servers. Right now.

Workflow

What's running

8 active systems

Hermes Personal AI orchestrator — voice, skills, persistent memory, self-evolving config
OpenClaw AI agent platform on Telegram — 53 skills, voice + STT, multi-model routing
WARD Open standard for AI bot access control on websites — getward.org, MIT
claude-code-security-hooks Prompt injection defense for AI coding agents — 31K+ impressions in 48h
Slavka memory pattern Persistent semantic memory for AI agents across sessions — open source
Vanga AI decision engine for market analysis, predicate-based reasoning
WebPrüfer AI-powered German legal compliance scanner — 80+ checks in 30s
n8n + RAG stack Self-hosted RAG: Qdrant + Voyage embeddings + Anthropic generation

See all projects

How it works

Three things I do, running in production right now

Memory

AI agents that remember

Hermes, OpenClaw, and the Slavka memory pattern give agents persistent context across sessions. Not 'chat history' — structured semantic memory with selective recall.

hermes.live

Live

[memory_recall] ✓ 14 memories loaded

[skill] crm_upsert → vaclavnew/crm:partners/marko.md

[llm_call] claude-sonnet-4-6 ✓ 1.4s · 247 tokens

[memory_store] ✓ project_marko_followup_2026_05_11.md

RAG

Retrieval-augmented Claude over my own corpus

The Ask Slava widget on this site is a working RAG: pgvector on my Postgres, OpenAI embeddings, Claude Sonnet 4.6 with prompt caching, citations back to the source project or post. Self-hosted, no SaaS retrieval layer.

ask-slava.live

Live

[query] "how do you handle AI memory?"

[embed] text-embedding-3-small ✓ 1536d · 84ms

[pgvector] top-K=5 → slavka-memory:2, hermes:1, openclaw:1, writing:1

[llm] claude-sonnet-4-6 ✓ cache hit · 142ms TTFB

[stream] 312 tokens · 2.1s · cited 3 sources

Compliance

German regulatory hell, turned into tooling

BFSG, DSGVO, Impressum, Datenschutz — laws that paralyze most agencies. WebPrüfer scans 80+ rules in 30 seconds and opens GitHub issues on what's broken.

webpruefer.live

Live

[scan] example.de → 80 checks queued

[BFSG] contrast ratio ✗ 2.8:1 (need 4.5:1)

[DSGVO] Google Fonts CDN ✗ external load

[gh_issue] 3 issues opened in repo

✓ done · 27s

Newsletter