Proof draft · LIVE=NO · custom domain pending CEO CF door
Mohamed A M Elansary, PhD
Target: Research Engineer, Domain Scaling — Anthropic
Sourced insights (≤2)
- Domain Scaling charter: Make Claude world-class at real-world knowledge work (finance, healthcare, legal) by owning end-to-end RL environments — high-value tasks, reward signals, vendor relationships, and measured model impact. Source: job-boards.greenhouse.io/anthropic/jobs/5271380008
- Reward hacking in real Claude RL envs: When models learn to reward hack during training in real RL environments used in Claude training, that correlates with increased misaligned behavior across evaluations — why Domain Scaling needs QA frameworks that catch reward hacking and protect env quality. Source: anthropic.com/research/emergent-misalignment-reward-hacking
Proof — domain data + UQ-as-eval + ship
- Domain RL envs / domain data (adjacent): Env Eng PhD + environmental / regulatory / sustainability knowledge work — read messy multi-source datasets, define success, spot quality issues. Honest: not Anthropic-scale RL ownership.
- UQ as eval: TAMUK 2022 dissertation — multi-basin hydrologic forecast uncertainty reduction (USGS/NOAA/NASA; HPC MODFLOW/VIC/PIHM/NASA LIS) as failure-mode measurement and generalization habit for reward/env QA.
- Ship: Production multi-tenant agentic LLM / RAG-adjacent systems (Claude, GPT, Gemini) — retrieval, query routing, isolation; WhatsApp AI receptionist + voice booking agents (Vertexium).
- Brand: the PhD who ships. Prefer take-home / work-sample when interview format allows. Prefer Seattle among Anthropic hubs.