Production Systems
Systems
Selected production systems — each with a real problem, a real build, and a real outcome.
ISNAD
The ProblemWhen a multi-agent AI system answers you, that claim has passed through many hands — a source, a scraper, an ingestion model, a synthesis model. Each hand can drop, distort, or invent. Provenance tools record what happened. Nothing grades who did it.
BuiltOpen-source, claim-level provenance framework adapted from classical Islamic hadith transmission science. Reference implementation with 1,000+ tests, five pluggable strategy interfaces, end-to-end worked example. Published paper (arXiv:2607.24117), PyPI package, LangChain integration.
OutcomeAdapts hadith narrator-grading and chain-level provenance to multi-agent AI pipelines. Validated on 20,000 claims; 4,057 quarantined claims traceable to specific narrator grades. Open-source on PyPI; product site at isnadhq.com.
Stack Python, multi-agent provenance, narrator grading, Bayesian grading, LangChain integration
Islam & AI
The ProblemGlobal users needed reliable, multilingual access to Islamic knowledge at scale.
BuiltIslamic knowledge platform — live today as a focused LLM assistant on a rebuilt DeepSeek backend, with paid tiers, an API dashboard, and Android live on Google Play, iOS in review. Qur'an + 600k+ Hadith knowledge base, memory, and deep research — live. Began as a Microsoft Imagine Cup participant.
OutcomeServing 145,000+ lifetime users across 190+ countries. Free core with paid tiers (Lite/Pro/Benefactor), an API dashboard, and Android app live on Google Play. Declined two acquisition offers (one six-figure, one seven-figure USD).
Stack LLM, knowledge base, multilingual NLP, OCR, memory
Full Story → · islamandai.com →Entropic Technologies
The ProblemTeams needed production AI execution, not fragile one-off automations.
BuiltGPT-powered email generation, document parsing with structured extraction, and memory-safe therapy platform APIs across logistics, healthcare, marketing, and data platforms.
OutcomeProduction systems shipped across industries — 18+ clients and $100k+ revenue — before the venture wound down.
Stack LLM orchestration, backend APIs, structured extraction, safety systems
Glacis — Voice AI
The ProblemLogistics teams were bottlenecked by manual dispatcher calls and follow-ups.
BuiltVoice AI agents for outbound calls, scheduling, ETA updates, and compliance-critical communication.
OutcomeDispatcher-heavy operations moved to live voice automation in production logistics flows — roughly 30–50% less operational workload.
Stack Voice AI, telephony integrations, workflow orchestration
MDVoice
Linkmdvoice.ai →
The ProblemClinicians were losing time on manual documentation after patient conversations.
BuiltReal-time doctor-patient transcription with SOAP note generation workflows.
OutcomeProduction-ready transcription and SOAP support for high-volume clinical documentation.
Stack Real-time transcription, clinical NLP, structured note generation
Maverick AI
The ProblemPublic safety and compliance needed automated computer vision during COVID-era operations.
BuiltComputer vision for public safety — mask detection, people counting, social-distance checking. Led AI R&D as Co-founder & Chief AI Officer.
OutcomeGrew engineering team from 3 to 12. Built computer-vision systems for malls across Saudi Arabia; the system never reached deployment.
Stack Computer vision, PyTorch, edge deployment, real-time inference
Genie AI / Albis
The ProblemTeams needed reliable speech and vision systems for real-world use cases.
BuiltBuilt the first MVP of Albis, a personalised 3D avatar assistant with TTS/STT voice interaction. Machine Learning Lead.
OutcomeShipped TTS/STT voice assistant systems and computer vision workflows for production use.
Stack TTS/STT, computer vision, OCR pipelines, 3D avatar rendering
Motive (KeepTruckin)
The ProblemFleet management needed automated AI testing pipelines for safety-critical systems.
BuiltProduction AI testing pipelines and systems behind thousands of vehicles and drivers. Landed via a $200 freelance project that went viral on LinkedIn — 726 reactions and six interviews.
OutcomeNegotiated the package up. Worked on production systems serving thousands of drivers.
Stack Production AI testing, fleet management systems, safety pipelines
Isharay
The ProblemAccessibility for the deaf and hard-of-hearing community in Pakistan.
BuiltPakistani Sign Language translator — computer vision + NLP. Runner-up at the Pakistan Science Fair Award (Ulster University), January 2020.
Outcome50% tuition scholarship toward Ulster MSc in Data Science, AI, or IoT. Judged by panel chaired by Ulster's Associate Dean for Global Engagement.
Stack Computer vision, NLP, accessibility
Facebook → Ulster University →BunBites.pk
The ProblemA real restaurant in Wah Cantt needed online ordering and a proper point-of-sale — without a SaaS subscription eating the margin.
BuiltLive online menu, WhatsApp ordering, and a full POS with kitchen display, inventory, and a dashboard — built on URY over Frappe, self-hosted, Cloudflare in front.
OutcomeLive at bunbites.pk, first month free for the restaurant. The design partner for a white-labelled restaurant platform I'm rolling out to other restaurants.
Stack Frappe/URY, POS, inventory, WhatsApp ordering, Cloudflare
bunbites.pk →Research Distribution as a System
The ProblemIndependent research dies in obscurity. No lab, no press office, no coauthor network.
BuiltA repeatable launch sequence — preprint and DOI, archived software release, package distribution, aggregator submission, community seeding, long-form technical writing, conference submission — each stage instrumented.
OutcomeWithin three weeks: an unsolicited technical comparison published by Paul Hammant, co-creator of Selenium, positioning his own project relative to the framework and proposing an interop path. Outside contributors filed issues and pull requests, including a code-level critique that surfaced an error in the paper's own published evaluation. Indexed on arXiv and Hugging Face Daily Papers.
Stack arXiv · Zenodo · PyPI · Hugging Face · Reddit · Medium · LinkedIn
Full breakdown: "I Have No Lab, No Press Office, and No Co-authors" →
It is the receipt.
Selective engagements for teams with real budgets and expensive workflow problems.