Igor Ganapolsky IgorGanapolsky

Igor Ganapolsky

I build AI systems, mobile products, and developer tooling.

Current focus:

ThumbGate.ai - self-improving agent governance that turns thumbs-up/down feedback into pre-action gates that block repeat AI mistakes
Random-Timer - native iOS and Android reaction timer for combat sports, BJJ, boxing, and HIIT
openclaw-console - native mobile console for monitoring and approving agent actions
hermes-eval - open-source multi-host LLM evaluation harness and quantitative 23-fold Walk-Forward + 2,000-simulation Monte Carlo benchmark suite

I use Zo Computer as a personal cloud computer for AI-assisted work with files, connected tools, automations, and hosted projects.

Referral disclosure: I receive $10 cash if you sign up and become a paying user; you receive $10 in Zo credits.

Project	What it does
ThumbGate.ai	Governance layer for AI agents: user feedback becomes enforcement that prevents repeated mistakes
Random-Timer	Mobile training timer built for combat sports and interval work
openclaw-console	Native mobile console for monitoring and approving agent actions
hermes-eval	Multi-host LLM evaluation harness, Golden-set CI gate, and quantitative 23-fold Walk-Forward + 2,000-sim Monte Carlo benchmark suite

Local execution & RTH surveillance: Apple Silicon local Python runtime (scripts/local_trading_sentinel.py).
Batch Walk-Forward, Monte Carlo & Context Memory: Tencent Cloud Lighthouse / CVM (SA5 / SA9 AMD Turin-Dense 3.4 GHz all-core turbo in ap-hongkong & ap-singapore) and Tencent Hunyuan (Hy-Memory-30B-A3B surgical state-patching pattern and OpenAI-compatible gateway).
Published Dual-Region SA9 & Hunyuan Case Study & Benchmark: hermes-eval#cloud--model-infrastructure · Full Technical Case Study (148.2 ms P50 in ap-hongkong / 151.6 ms P50 in ap-singapore, -46.3% cost vs AWS c7a, -78.42% context token burn via Hunyuan Hy-Memory).