Danil Avakumov
@malishomen
Independent AI-safety researcher. Solo, 18 months. Built BDC Bridge — a live bounded-evaluation service for agentic AI workflows (bridge.bdc-hive.com). 34 measured gates, 32 PASS / 2 preserved FAIL. Preserved-failure discipline: R10 and R15 stay canonical, not silently rewritten.
https://bridge.bdc-hive.com$0total balance
$0charity balance
$0cash balance
$0 in pending offers
About Me
BDC Bridge is the production face of an 18-month independent AI-safety research line. Six bounded mechanisms cooperatively assembled. 34 measured gates total — 32 passed, 2 preserved as canonical failures. External chaotic-validation experiment: 6 of 12 architectural laws confirmed across 2 LLM providers and 2 languages on real-world chunks. Calibration: Tier 2 PASS at the operator-controlled tier; Tier 3 needs 10+ admissible third-party deployment cases — exactly the gap this Manifund grant is asked to close.
Live service: https://bridge.bdc-hive.com
Public BDC overview: https://bdc-hive.com
User-help repo (public, MIT): https://github.com/malishomen/bdc-bridge-passport
GitHub: https://github.com/malishomen
LinkedIn: https://www.linkedin.com/in/danil-avakumov
Discipline: preserved-failure rule, no silent rewrites. R10 and R15 stay canonical FAILs in the canonical record.