Evidence
Every claim on this site links to where it came from.
No adjectives stand in for proof. Here are the artifacts you can open yourself.
Open any of these
Hadith benchmark · κ = 0.87
The weakest-link rule run against 575,060 scholar-graded chains. κ = 0.8714 (3-way; 1 = perfect, 0 = chance) — agreement with a rule-based convention (Ibn Hajar's 12 tiers), not ground truth. 5-way 0.8667, lenient 0.761, narrator-grade κ = 0.33, shuffled control ≈ 0 (-0.0066). 577,024 is the raw corpus of sanads — the benchmark N is 575,060.
Results in the repo →RAGTruth transfer · κ = 0.4345
1,800 responses from 6 models: 96.0% hallucination recall at 60.4% precision (recall ≠ precision), 70.3% accuracy against a 55.8% baseline. 96.8% parsed; the 3.2% that did not parse are scored fail-closed, not excluded.
Read case #0 →What evidence actually means
AI system inventory — every model, agent, and pipeline in use, versioned and located.
Risk classification — which systems are high-risk, under which framework, and why.
Decision and event logging — what the system did, when, and on what input.
Provenance and lineage — where each claim came from, through whom, and whether the chain holds.
Human-oversight evidence — who reviewed what, and the trail that proves it.
Technical documentation pack — the artifacts an auditor can actually open.
Deadlines are already set
Annex III high-risk obligations 2 December 2027; Annex I 2 August 2028 (moved by the Digital Omnibus, in force 27 July 2026).
Art. 50 transparency duties for providers and deployers have applied since 2 August 2026.
Art. 12: high-risk AI systems — automatic event logging (record-keeping) · Art. 19: providers of high-risk AI systems — logs kept at least six months · Art. 18: providers of high-risk AI systems — documentation kept 10 years.
SDAIA published its AI Adoption Framework 12 September 2024 — in practice a procurement filter for government entities adopting AI (model accountability, transparency, human oversight) — and a National AI Risk Management Framework on 14 July 2026.
UAE Central Bank: Guidance Note on the Consumer Protection and Responsible Adoption and Use of Artificial Intelligence and Machine Learning by Licensed Financial Institutions in the U.A.E. (issued 11 February 2026).
The delay is not relief. Evidence architecture takes months to build, and the technical standards are still being drafted. That gap is the work.
Open source, already running
ISNAD
What happened, through whom, and which link was weakest. Listed in LangChain's middleware integrations.
ISNAD →The way in
Free 30-minute call
For qualified teams.
We check the four-question filter and look at your pipeline. No deck, no pressure.
Discuss the call →Provenance Assessment
two weeks · one pipeline · $5,000 (first 2) · $7,500 after
A technical readiness review with a written report, signed evidence, and a coverage map.
See the process →Annual licence & ongoing
For teams that want it to keep running.
ISNAD in your pipeline, the evidence kept current, and me on call for the hard questions.
Let's talk →A technical readiness review — not an Article 43 conformity assessment and not a notified body.
Trusted by people I've built with
"I'm happy to recommend Ali, who worked with us as a contractor over the past four months on an MVP for an LLM-based knowledge base. Ali was friendly, approachable, and a pleasure to work with. He took constructive feedback well, listened carefully, and adapted his approach as the project developed. When things got challenging, he stayed engaged and pushed through rather than stepping back. We also appreciated his initiative beyond the core project. Ali brought forward ideas for the company and took on additional work to help put them into practice. He was willing to contribute beyond his original scope and help wherever he could add value. I'd gladly recommend Ali to a team looking for an adaptable, proactive colleague who responds well to feedback and keeps working through challenges. I wish him every success in his next role."
Steffen Höhne
Founder, SHCV.IT OÜ
"Ali is an exceptionally talented software engineer with deep expertise in end-to-end system design and a passion for advancing AI and ML technologies. His ability to design and build AI systems is outstanding."
Noah Marra
Sr. Machine Learning Engineer, AMD · Former teammate
"Ali Raja is an exceptional and talented data scientist that possesses great entrepreneurial and technical abilities. Working with him is always a pleasure."
Ali Almussa
Entrepreneur & Project Management Consultant