Ali Zahid Raja · Applied AI Engineer & Architect
AI systems that survive production.
I take customer problems from discovery to production. I help teams find the right technical approach, design the system, build the integrations, and get AI working inside real workflows.
Available full-time, as a fractional CTO, or forward-deployed with your team.
Pick the one that fits where you are
A full-time role, a fraction of my time, forward-deployed with your team, or a focused two-week assessment.
Full-time
Applied AI Engineer or Architect
Remote from Pakistan, comfortable overlapping with EU and US hours.
Talk to me →Forward-deployed
Embedded with your team
I embed with your team and ship the thing: knowledge bases, agents and local LLM deployments, inside your own infrastructure.
Talk to me →Fractional CTO
2–3 days a week
For pre-seed and seed teams that have customers and code but no technical lead: architecture, roadmap, first hires, engineering leadership, and the technical side of your customer, procurement and standards conversations.
I've been CTO of a business I founded (18+ clients, six figures), grown a team from 3 to 12 as a co-founder, and led EU-facing standards work (EU AI Pact pledge, EU AI Office correspondence, UK DSIT submission).
Starts as a 3-month engagement, with 2 weeks' notice either side.
Talk to me →Provenance Assessment
$5,000 for the first 2 design partners, $7,500 after
Two weeks, fixed price. I trace how your AI builds its answers and show you where they can't be trusted. Built on ISNAD.
See the two-week process →A technical readiness review — not an Article 43 conformity assessment and not a notified body.
Every number links to where it came from
The two results that matter, with the caveats printed next to them.
Hadith benchmark · κ = 0.87
The ISNAD grading rule run against 575,060 scholar-graded hadith chains. κ = 0.8714 (1 = perfect, 0 = chance) — agreement with a rule-based convention (Ibn Hajar's 12 tiers), not ground truth. Inter-scholar agreement on the same material sits at κ = 0.33. The 577,024 figure is the raw corpus, not the benchmark N.
Full breakdown →RAGTruth transfer · κ = 0.4345
On 1,800 responses from 6 models: 96.0% hallucination recall at 60.4% precision (recall ≠ precision). 96.8% parsed; the 3.2% that did not parse are scored fail-closed, not excluded. Read as the honest, not the flattering, number.
Read case #0 →From customer problem to production system
Claim-level provenance
ISNAD grades every source, scraper, model and agent behind an answer, caps the chain at its weakest link, and signs the record — adapted from 1,200 years of hadith transmission science.
Explore ISNAD →Evidence architecture
The logging, provenance and documentation an auditor can actually open — inventory, risk classification, decision logs, and the human-oversight trail.
See the evidence layer →Production AI systems
Voice AI, knowledge systems, backend orchestration and workflow automation that survive real users and real load — built for teams with expensive problems.
View the systems →Research & writing
Independent, single-author research — published, benchmarked, and written up so practitioners can check the work instead of taking my word for it.
Read the writing →Work that ships
ISNAD
BuiltOpen-source, claim-level provenance framework. 1,000+ tests, published paper (arXiv:2607.24117), PyPI package v3.2.0.
Explore ISNAD →Gnosis
BuiltSigned provenance for LLM knowledge bases: a website-to-Markdown crawler with SHA-256 hashes, WARC archival and Ed25519 signatures, so every file proves where its bytes came from. Built at SHCV.IT (MIT, v2.4.3 on PyPI); its capture-record spec drew a reply from the EU AI Office and now tracks CEN/CENELEC prEN 18284.
GitHub →Entropic
OutcomeAn AI Operating Brain for founder-led agencies — 18+ clients and $100k+ revenue before it wound down.
See the record →Glacis — Voice AI
OutcomeDispatcher-heavy logistics flows moved to live voice automation in production.
All systems →Customs & sanctions screening
Outcome5,800+ pages of official sources and 55,000+ sanctions entries refreshed daily — piloted as a Microsoft Teams assistant inside a client's own tenant.
All systems →From people I've worked with
"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
"Really enjoyed the session today. It was very casual and easy to follow, but at the same time packed with practical and informative insights. Loved how you shared your own experience with the agents, and I genuinely took away a few things I'll be trying myself — like automating my code base for testing."
Danish Elahi
Topmate session · Sep 2026
"Thank you for explaining your AI work model to us. You made a great point about not getting caught up in every trendy AI term or tool; instead, we should concentrate on what works for our specific use case."
Anees Iqbal
Topmate session · Sep 2026
4.8/5 · 106 ratings · 565+ sessions · 90 testimonials on Topmate
Beyond engineering engagements
Advisory
Positioning, strategy and research-distribution for technical founders and researchers who've built something real and can't get it seen.
Learn more →Global Tech Leverage
A private network for experienced engineers moving from local markets to global remote roles. “Talent is global. Opportunity is not.”
Learn more →Let's find the right technical approach for your problem.
Tell me what you're trying to ship. If it's not a fit, I'll say so on the call.