I'm a product-facing full-stack AI developer at KUEC. These days, I'm also building Notch, and running Socia, on a mission to curate the most active builder community in the UAE.
Away from work, I play table tennis, vlog my experiences, and occasionally direct short films. I also love hackathons and now host them under Socia.
A workspace for meetings, presentations, and demos: a voice-following teleprompter plus recording, editing, and recaps, with 1,500+ trial downloads and 100+ paying users.
Started as an offline, voice-following teleprompter (Electron, Vosk, TypeScript) and is growing into one connected workspace to prepare for, deliver, record, and revisit meetings, presentations, and demos.
The teleprompter stays the entry point. Writer turns rough notes into a script, screen recording captures the moment, a built-in editor handles auto-zoom, captions, and cuts, and Recap brings back what was said afterwards.
Privacy-first by design: voice-following and processing run locally, with no accounts or cloud dependency.
Built for people whose work is calls and follow-up, such as sales and account teams, customer success, educators, consultants, and team leads, and shaped by 1:1 customer discovery on positioning, UX, pricing, and roadmap.
Launched at a 24-hour virtual sprint run through Socia, then scaled to 1,500+ trial downloads and 100+ paying users.
A national AI platform that carries university research from invention disclosure to commercialization for five stakeholder groups, now in pilot UAT with Khalifa University and MOHESR.
Leading a six-person team building a unified platform that moves research from idea and invention disclosure through IP review, industry collaboration, investment, and commercialization for five stakeholder groups.
Own end-to-end product and technical delivery across architecture, development, QA, stakeholder discovery, pilot deployments, and demos, including conversations on multi-university adoption.
Built AI capabilities for document intelligence, semantic search, prior-art support, missing-information detection, research-to-partner matching, and decision-ready reports using retrieval-augmented generation and vector search.
Now in pilot-stage user acceptance testing with Khalifa University and MOHESR.
An autonomous data scientist built for the MBZUAI K2 Think hackathon: give it a dataset and a five-stage agentic pipeline plans, codes, tunes, and evaluates a model in under a minute, with reproducible code behind every result.
Developed for the MBZUAI K2 Think hackathon, designing a five-stage agentic workflow (EDA, Plan, Synthesize, Train, Evaluate) coordinated by the K2-Think model.
Optimized the full pipeline to under a minute, including LLM reasoning, a 20-trial Optuna hyperparameter search, and model training.
Built a deterministic code-synthesis engine on Jinja2 templates so every generated ML script is reproducible and auditable.
Streams progress to a Next.js interface over SSE and Redis, and produces an evaluation report and a business dashboard from every run.
FFillyComing soon
A review-first Chrome extension that plans an entire job application in one pass, grounds every answer in your own evidence, and never clicks submit.
Matches exact profile values in code, then plans the whole form in one batch as typed actions: text, dropdown, radio, checkbox, file upload, skip, or needs review.
Resolves ambiguous fields against saved evidence and the options actually available on the page, and withholds any answer it cannot support.
Keeps the profile and documents local in IndexedDB, parses resumes into the profile, and never clicks final submit.
Ships with Meaning Find, a companion extension that searches a page by meaning and highlights the original passages.
Covered by 29 automated tests across evidence selection, uncertain choices, and relay security.
A retrieval-augmented pipeline that turns dense railway legal documents into question answering and BPMN process models, built during my AI lab internship at Ab Ovo.
Built a RAG pipeline that converts legal documents into real-time B2B insight using vector databases and prompt engineering.
Implemented semantic chunking and embedding pipelines, testing Hugging Face models to improve retrieval quality.
Experimented with instruction tuning and LoRA fine-tuning on Llama 3 variants for railway-domain fidelity.
CDPCF2026
A cultural data provenance and compliance framework that protects intangible heritage from unauthorized AI training, placing 2nd at the UN Global SDG Public Policy Innovation Challenge out of 866 teams.
Designed a framework for tracing the provenance of cultural data and checking its use against compliance rules, so intangible heritage is not used in AI training without authorization.
Took 2nd place at the UN Global SDG Public Policy Innovation Challenge, representing the UAE in China as the Middle East’s only finalist out of 866 teams.
Building AI-native solutions across Khalifa University Enterprises Company, leading product and development from UVenture to internal tooling and client engagements.
Lead product and development for AI-native solutions at KUEC, owning the path from discovery and architecture to build, QA, pilot, and demo.
Flagship work is UVenture, a national research commercialization platform built with a six-person team and now in pilot-stage UAT with Khalifa University and MOHESR.
Build internal tooling that automates and augments the team’s own workflows, and take on client engagements end to end, from requirements through delivery.
Run stakeholder discovery and turn requirements into product, including discussions on multi-university adoption.
Build the AI layer behind these products: document intelligence, semantic search, prior-art support, research-to-partner matching, and decision-ready reports using RAG and vector search.
Co-founded and scaled a UAE-based hackathon community and builder network to 1,000+ builders across 70+ colleges, leading partnerships, programs, and events.
Scaled the UAE’s fastest-growing hackathon network to 1,000+ builders across all seven emirates and 100+ institutions.
Closed and manage delivery of an eight-month paid engagement with the National MS Society UAE, building a letter-of-intent and grant management system into their legacy workflows.
Secured partnerships with Replit, 42 Abu Dhabi, and Odoo Middle East DMCC, running discovery calls, proposals, and design.
Ran a five-day Odoo bootcamp, a 250-application buildathon with 42 Abu Dhabi and Replit, and the 24-hour virtual sprint that launched Notch.
