Chief AI Officer @ Kanban Studios — I design multi-agent AI systems and ship them as real products.

Agentic workflows · RAG & document intelligence · govtech decision engines · voice AI · industrial automation — built end-to-end, deployed, and auditable.

I'm a Software Engineering student in Abu Dhabi who builds AI products the whole way through — data, models, agents, backend, frontend, deployment — not just notebooks.

• 🧠 Chief AI Officer @ Kanban Studios — leading the AI product arm: a productized AI-service catalog, RAG & document-intelligence workflows, the Kanban Agents autonomous-automation infrastructure, AI-safety frameworks, and Stripe-ready SaaS.
• 🔬 Visiting Researcher — Mixed Reality & Automation at BRIDGE (EDGE Group), working on an Industry 4.0/5.0 smart-warehouse platform: real-time HMIs, a Three.js digital twin, sensor-connected Raspberry Pis, and ML forecasting.
• 🚀 Founder & CTO of XPBridge — a mobile app I designed, built, and shipped solo to the Google Play Store.
• 📄 Undergraduate Researcher @ Al Ain University — published my first IEEE paper at IEEE REW 2026: explainable requirement-defect screening, tested across 79+ SRS documents and 9,000+ requirement statements.
• 🤖 My specialty is agentic AI that can be trusted: multi-agent pipelines with deterministic guardrails, compliance critics that can veto, fairness checks, and immutable audit logs — the boring parts done right.
• 🎓 BSc Software Engineering @ Al Ain University (expected 2027) — CGPA 3.73 / 4.00, named to the Honor List three consecutive semesters.
• 🏆 9× competition podiums & honors across UAE tech events — 4× 1st place (incl. G42's Tamkeen 5.0 Agentic AI track, the University Track at INSPIRE '26, and 1st of 48 teams at Oriane × Replit's Build for the Video Economy), 2× 2nd, 1× 3rd, plus finalist & championship recognition.
• 🌍 I work in English, Urdu, Hindi and Pashto (fluent), plus basic Arabic — and I ship bilingual EN/AR interfaces by default.

Languages

AI / ML

Frontend & Mobile

Backend & Data

Cloud · DevOps · Hardware

Tools & Design

• 🧠 Building Kanban Studios' core AI capability as Chief AI Officer — shaping the AI service catalog, next-gen RAG & document-intelligence demos, the Kanban Agents infrastructure, AI-safety frameworks, and Stripe-ready SaaS products.
• 🏭 Advancing the ASRS platform and industrial-automation R&D as a Visiting Researcher at BRIDGE (EDGE Group) — real-time sensor connectivity, machine communication, and smart-automation systems.
• 🧠 Going deeper on agent architectures: LangGraph state machines, tool subgraphs, and the Model Context Protocol (4 Anthropic certifications, May 2026).
• 📱 Growing XPBridge post-launch — new AI features, retention, and store performance.
• 🎓 On track to graduate in 2027; aiming at agentic-AI and full-stack engineering roles where systems must be fast, bilingual, and auditable.

🧠 Kanban Studios — Chief AI Officer (CAIO) (Jul 2026 – Present) — leading AI strategy and engineering for the studio:

• Productizing AI services — a robust AI service catalog, from bespoke automation systems to intelligent business tools.
• Next-gen AI demos — cutting-edge Retrieval-Augmented Generation (RAG) models and advanced document-intelligence workflows.
• Agentic workflows — enhancing the Kanban Agents infrastructure for autonomous, high-efficiency business automation.
• AI safety & risk mitigation — frameworks that keep every deployed model secure, compliant, and reliable.
• Stripe-ready SaaS — scalable commercial software products built for global monetization.

🔬 BRIDGE — EDGE Group PJSC · Abu Dhabi · 10 months, 3 progressive roles

• Visiting Researcher — Mixed Reality & Automation (Jan 2026 – Present) — ongoing R&D on the ASRS platform, real-time sensor connectivity, and industrial machine communication.
• Intelligent Systems & Automation Trainee (Sep – Dec 2025) — ASRS 2.0 smart-warehouse platform: shipped 5+ Svelte/TypeScript HMI pages (EN/AR, dark/light), wired them to 5+ FastAPI endpoints, built a unified Three.js digital twin with live motion logic, integrated 2 Raspberry Pis over WebSockets for safety stops, implemented mTLS + RBAC + rate limiting, built an XGBoost sales-forecasting model, and resolved 50+ bugs in weekly Agile sprints. Full internship report →
• Software Engineer Intern (Aug – Sep 2025) — programmed the Boston Dynamics Spot robot for border-patrol use cases; co-designed ASRS's 3-tier architecture from stakeholder requirements.

🚀 XPBridge — Founder & CTO (Jan 2026 – Present) — own the full stack of a published mobile product: React Native frontend, Supabase/PostgreSQL backend + auth, AI features, and the complete lifecycle through Google Play release.

📄 Al Ain University — Undergraduate Researcher (Jan 2026 – Present) — published my first IEEE paper, in Requirements Engineering:

• Keeping Engineers in the Loop: Explainable Requirement Defect Screening Across Software Requirements Specifications — 2026 IEEE 34th International Requirements Engineering Conference Workshops (REW).
• Explainable, rule-based analysis of software-requirements quality, tested across 79+ SRS documents and 9,000+ requirement statements.
• Owned it end to end: research design, analysis, writing, and presenting.

🎓 Al Ain University — BSc Software Engineering (expected 2027) — CGPA 3.73/4.00, Honors List, Abu Dhabi campus.

Compresses the UAE Ministry of Energy & Infrastructure's official 5-working-day housing-arrears review into a sub-10-second, fully auditable AI decision.

• Problem → manual officer review of rescheduling requests against a 6-rule governance rulebook.
• Solution → a LangGraph StateGraph of 11 agents: PDF salary-certificate parsing with LLM field extraction, deterministic governance rules (G-01→G-06), a Compliance Critic subgraph that can veto and force escalation, a fairness check against historical decisions, bilingual EN/AR rationale, WhatsApp notification, and immutable audit logging.
• Built to scale → service-agnostic engine, proven by a second federal service (visa renewal) running through the same pipeline — validated by 51 passing unit tests.
• ⚙️ Next.js 16 TypeScript LangGraph LangChain Groq · Llama 3.3 70B Supabase (Postgres + RLS) pdfjs-dist Twilio
• 💡 Why it matters: a working template for the UAE Federal Government's directive that AI handle 50% of services — with the audit trail regulators actually require.

A follow-up loop for UAE hospitals that patients never have to install: their care plan, a check-in every night, and answers on WhatsApp in their own language, with a nurse alerted the moment a warning sign appears.

• Flow → a nurse uploads the discharge letter; Gemini 2.5 Flash reads it into a care plan the nurse approves; the patient receives it on WhatsApp in one of five languages. Questions are answered strictly from their own discharge instructions.
• Safety → symptom reports and voice notes are triaged against the patient's own warning signs. Red means a critical alert and an urgent reply, and emergency keywords escalate instantly. If the model is down, a reported symptom is still acknowledged and escalated.
• Engineering → multi-tenant Supabase (Postgres 17 + Row Level Security), Vercel Cron + pg_cron scheduling, and 11 check suites, including 139 Arabic phrasings for the intent classifier and an end-to-end webhook test against an in-memory Supabase.
• ⚙️ Next.js 16 React 19 TypeScript Supabase Twilio WhatsApp Gemini 2.5 Flash Vercel
• 💡 Why it matters: healthcare AI has to fail safe. Every path, including a model outage, ends with a human seeing the warning sign.

Driver-fatigue detection from a plain webcam or phone. A CNN trained from scratch and facial geometry are fused over time into PERCLOS, the automotive fatigue standard, so ordinary blinks never set off the alarm.

• Model → a 139k-parameter CNN trained on 85k infrared eye images with subject-wise splits (no data leakage): 98.66% accuracy and 0.997 ROC-AUC on 6 people it has never seen. The threshold is chosen for recall, and the weaker subgroup (glasses, 95.35%) is reported, not hidden.
• System → MediaPipe landmarks (eye and mouth openness, head pose) are fused with the CNN in a state machine covered by 70 simulated-time tests. Adaptive night and sun correction raises lighting robustness from 5/7 to 7/7 conditions.
• On device → a native Android port (Kotlin + ONNX) that requests only CAMERA and VIBRATE, so "runs entirely on the phone" is enforced by the OS. It has its own 22-test JVM suite and a parity check against the Python version.
• ⚙️ PyTorch OpenCV MediaPipe ONNX Kotlin Android
• 💡 Why it matters: honest evaluation. A random split would have scored higher and then failed on a real face.

A money app for parents in the UAE, built with a teammate at a hackathon. Spending, goals, loans and kids' chores live on the phone, with Gemini-powered helpers on top.

• AI features → a streaming chat coach that answers from the family's own numbers; receipt scanning with Gemini vision; a "should I buy this?" check that runs a purchase verdict and a deal finder in parallel; and a next-month forecast grounded in Google Search, where the model proposes percentages and the server does the arithmetic.
• Engineering → a pnpm monorepo (Expo app + Express 5 API) built around one OpenAPI contract. Orval generates both the typed client and the server's Zod request validation. When the AI is unavailable, the coach falls back to rules and the forecast to three-month averages.
• ⚙️ React Native (Expo) expo-router Express Gemini 2.5 Flash OpenAPI + Orval TypeScript monorepo
• 💡 Why it matters: the family's data stays on the device (AsyncStorage). Only a receipt photo or a text summary leaves it, and only when an AI feature is used.

Fully hands-free voice loop: speech-to-text → LLM reasoning → spoken reply, in real time.

• Pipeline → Groq Whisper for low-latency transcription, Gemini for conversational reasoning, browser SpeechSynthesis for natural playback.
• ⚙️ Next.js FastAPI Groq Whisper Gemini API Web Audio API — deployed on Vercel + Hugging Face Spaces
• 💡 Why it matters: voice is the hardest UX to fake — latency, turn-taking, and audio plumbing all have to work end-to-end.

Answers university-specific questions with grounded, cited responses pulled from real documents — not hallucinated ones.

• Retrieval → FastEmbed (bge-small-en-v1.5) embeddings in Qdrant Cloud; the best-matching sections go to Llama 3.3 70B on Groq, and every answer returns the sources it used.
• ⚙️ React Vite FastAPI Qdrant FastEmbed Groq · Llama 3.3 70B — frontend on Vercel
• 💡 Why it matters: citation-grounded RAG is the pattern every serious org needs before trusting an LLM with domain questions.

Founded, built, and shipped solo — from first architecture decision to store approval and ongoing maintenance.

• ⚙️ React Native Supabase PostgreSQL AI integration Google Play
• 💡 Why it matters: shipping to a store — review process, crash-free sessions, real users — is a different sport from a demo repo.

9+ competition builds across agentic AI, edge AI & computer vision, creator-economy tools, safety tech, education, sustainability, and fintech — in front of some of the UAE's toughest judging panels.

More competition builds on this profile:

• 🗺️ Reach — AI siting copilot for Abu Dhabi: a "15-minute city" access map that finds under-served neighborhoods and uses an LLM to recommend a real, zoning-aware parcel for the missing facility, then simulates the improvement live. Next.js MapLibre H3 Groq
• 🌌 Aletheia — built for the Cognee "The Hangover Part AI" hackathon (Best Use of Open Source track): an AI research assistant that can unlearn — retract a source and watch the discredited knowledge visibly die out of its graph while answers re-derive. Cognee (self-hosted) Knowledge graphs

🤖 Agentic AI & RAG

• saddad → SADDAD — 11-agent govtech decision engine · live
• repeat → browser agent: show a workflow once, then press Tab to replay it, with one undo slider for the whole run
• mirage → Sybil resistance under AI-agent swarms: live coordination detection with evidence, plus correlation-weighted on-chain voting
• aletheia → Aletheia — research assistant with auditable unlearning
• AI-University-Assistant-RAG-Chatbot → AI University Assistant — cited-answer RAG chatbot (Qdrant + Groq) · live

🏥 Health AI

• CareLoop → WhatsApp post-discharge follow-up with triage and nurse escalation · live

👁️ Vision, Voice & Applied ML

• drowsy-driver-detection → real-time driver-fatigue detection: CNN + PERCLOS, 98.66% on unseen subjects, native Android port
• Jarvis — real-time voice assistant (Whisper → Gemini → TTS) · live
• Sentiment-Analysis → DistilBERT sentiment classifier (PyTorch) — model on Hugging Face · live

🌐 Full-Stack Web

• reach → Reach — 15-minute-city AI siting copilot for Abu Dhabi
• Fashion Store — responsive e-commerce demo (Next.js + Tailwind) · live

📱 Mobile

• XPBridge — founder-built app, live on Google Play
• parentwise → ParentWise — family money app for UAE parents with Gemini chat, receipt scanning and forecasts (Expo) · hackathon team build

🦾 Robotics & IoT

• spot-controller1 — Boston Dynamics Spot movement and docking scripts from my EDGE Group internship
• spot-digital-twin — rigged, animated Spot model for browser digital twins: 17 SDK-named joints, 7 animation clips, Three.js-ready

🔬 Research

• research — Multilingual Trust-Debt pilot: reproducible experiments on cross-lingual trust gaps in small open-weight LLMs

1. 🚀 Published release v1.0-neathack in Mohammad-Umar7/neathack-agent

_{⏱ auto-refreshes every 6 hours via GitHub Actions}

Mohammad-Umar7 · GitHub Ecosystem
│
├── 🤖 Agentic AI & RAG
│   ├── saddad                   → SADDAD · 11-agent government decision engine
│   ├── repeat                   → browser agent that learns from one demonstration
│   ├── mirage                   → detects coordinated AI-agent swarms, live
│   └── aletheia                 → Aletheia · the assistant that can unlearn
│
├── 🏥 Health AI
│   └── CareLoop                 → WhatsApp post-discharge follow-up · live
│
├── 👁️ Vision, Voice & Applied ML
│   ├── drowsy-driver-detection  → on-device driver-fatigue detection · Android
│   ├── Sentiment-Analysis       → DistilBERT classifier · weights on Hugging Face
│   └── (Jarvis)                 → live voice assistant · jarvis-ai-voice.vercel.app
│
├── 🌐 Full-Stack Web
│   └── reach                    → Reach · AI siting copilot for Abu Dhabi
│
├── 📱 Mobile
│   ├── (XPBridge)               → live on Google Play
│   └── parentwise               → ParentWise · Gemini money coach for parents
│
├── 🦾 Robotics & IoT
│   ├── spot-controller1         → Spot movement & docking scripts (EDGE Group)
│   └── spot-digital-twin        → rigged, animated Spot model for Three.js twins
│
└── 🔬 Research
    └── research                 → Multilingual Trust-Debt pilot (LLM evaluation)

I'm open to conversations with recruiters, founders, engineers, researchers, hackathon teammates, and open-source collaborators — especially around agentic AI, govtech, industrial automation, and products that need to ship, not just demo.