Johns Hopkins SAIS (MA 2026). AI governance and verification: I build the tools I use to check claims about AI systems, with deterministic verification built in, not bolted on. Currently a SPAR research mentee on verification evidence for compute governance.
overdue-ai: are frontier AI labs keeping the dated safety promises they made themselves? A source-cited board of individual commitments (RSP, Preparedness and Frontier Safety milestones, the Seoul and White House voluntary commitments) with live overdue and countdown timers and Wayback-archived evidence for every row. Live at overduetracker.org.
SPAR, Kairos AI Project (Fall 2026): structuring a sourced verification dataset on the project's evidence track: co-developing the coding scheme and piloting it on IAEA safeguards instruments.
| Project | What it does | Evidence |
|---|---|---|
| fieldatlas | Automated literature review for AI safety and AI policy: every claim verified against source text, every citation linted against a real corpus | Blind expert re-read of 15 papers: 15/15 faithful, 0 fabrications |
| ledger-datacenters | Verified-claim literature review of the political economy of the US AI data-center buildout | 2,071 claims through verbatim, numeric and entailment gates; coverage limits disclosed (draft) |
| nla-eval | Evaluation of the released Natural Language Autoencoder checkpoints on Qwen and Gemma, run on an H100 | Data and scripts on Zenodo; negative result reported as such |
| gpuwatch | Agent-safety study: purchase agent with hard-coded spend caps and an HMAC confirmation gate that config cannot override | 41 security tests passing |
| bluedot-tais-puzzle-solution | BlueDot Technical AI Safety Puzzle 1: identifying and characterizing a non-linear feature | Written solution |
| Project | What it does | Evidence |
|---|---|---|
| aslan-core | Bitemporal financial-data platform (Turkish market). Immutable history, content-addressed provenance, LLM extraction audit trail | 29K lines, 9 releases, ~1000 tests |
| emersusai | RAG health assistant over 1.1M biomedical papers. Refuses rather than hallucinates | Six weeks, Claude Code as coding agent (archived) |
| h1b-labor-map | Federal-data ETL: which employers sponsor H-1B, where, in which roles. Every threshold cited to the Federal Register | Ran on real multi-GB federal releases |
| dc-frontier-events-aggregator | DC AI and frontier-tech event aggregator. Multi-source dedup and relevance ranking | Live at events.emersus.ai |
Writing and CV: sidaraslanoglu.com. Turkish (native), English (C2). ORCID.