Hasan Kazmi

Independent AI-evaluation researcher and systems architect focused on benchmark methodology, source-grounded judgment, adversarial review, failure-mode analysis, and calibrated conclusions. Creator of AURORA GRID OS, a cross-model evaluation architecture for separating evidence from inference, testing claims against falsifiers and constraints, and preserving auditable decision records. Produces concise public-facing research and retains human judgment while using coding and AI agents as tools.

Founder & Decision Intelligence Systems Architect at AURORA GRID OS / GrindWire (2023 – Present)

• Designed a multi-stage evaluation architecture spanning signal intake, source verification, claim classification, constraint analysis, adversarial review, cross-model adjudication, confidence calibration, and auditable action states.
• Built K-ALIGN, BLACKGLASS, CRF, COMMAND, AAIK, and RECORD LOCK methods to detect unsupported claims, reasoning gaps, omitted evidence, narrative contamination, weak falsifiers, and overconfident conclusions.
• Created repeatable evaluation artifacts including source tiers, field cards, decision briefs, trigger maps, falsifier tests, quality-control checklists, correction records, and public methodology documentation.
• Developed and tested workflows across GPT, Gemini, Grok, and other agentic research tools, using models for retrieval and analysis support without delegating final judgment or source accountability.

Independent Technical Writer, Systems Auditor & Research Analyst at Freelance (2011 – Present)

• Produce evidence-led research briefs, technical audits, SOPs, runbooks, decision memos, source-chain analyses, and executive summaries across technology, strategic risk, operations, and public-record questions.
• Evaluate AI-generated outputs for factual accuracy, methodological validity, source quality, logical consistency, requirement alignment, uncertainty, and unsupported inference; convert findings into specific revisions and test criteria.
• Analyze dense and conflicting information, identify the binding constraint, distinguish verified records from interpretation, and communicate conclusions for both technical and nontechnical audiences.

Store Manager & Technical Account Specialist at AT&T Wireless Outlet (2010 – 2015)

• Led client-facing technical support, escalation handling, staff training, documentation, operational quality control, and performance tracking in a high-volume telecom environment.

– James Madison High School (2004 – 2008)