Seasoned professional with over 15 years of expertise spanning Machine Learning, Backend Engineering, HPC Kernel and Embedded Software Development. Skilled in delivering cutting-edge solutions across industries such as IoT, SaaS, Finance, Energy, Blockchain and Healthcare.
Proficient in a wide range of programming languages, having surpassed language-specific limitations through extensive experience and adaptability. As a fast learner with an entrepreneurial spirit, I bring critical thinking, mathematical insight, and exceptional communication skills to every challenge, driving innovation and achieving impactful results.
I bring a versatile skill set to every project. Whether it’s research or development, I’m all about delivering top-notch results that exceed expectations.
From POC to enterprise-level IT solution, I offer a range of solutions tailored to your needs. I’m here to help you make it happen.
Expert in AI SaaS, delivering scalable, custom AI solutions to drive business innovation: LLM, Energy, Finance, etc.
Building robust backends for small to enterprise applications, ensuring performance and robustness.
Expert in Embedded Systems, HPC, Kernel and Algorithms: delivering high-performance, efficient solutions.
Expert in ML innovation and POC development, turning data into actionable insights, from IoT to Game Agents.
Here’s a glimpse of the work I’ve done for clients and projects that I’m proud of. Each piece tells a story of creativity, strategy, and execution coming together to achieve success.
Xilos is a software platform for securing, orchestrating and refining your organization's AI queries through restriction & routing rules.
Your private, secret, and stealth copilot for all meetings. Say goodbye to stress and failure — Ntro.io is here to help you excel job interviews, professional meetings, client calls and even small talks without any language barrier!
Smart Smart City Expo Dubai brings together leaders from the best smart cities, futuristic innovative companies, governments and state-of-the-art organizations.
On-spot object detection/classification AI app for clinical labs with electronic microscope. It's trained on MICELL - The World's Largest Image Dataset for Microscopy Human Body Fluid Test.
Multi-state lottery game with MEGA jackpots and MEGA fun. You can play Mega Millions in 45 states plus the District of Columbia and the U.S. Virgin Islands; a total of 47 jurisdictions.
Stockfox is a virtual share-market advisory app that provides expert analysis, advice, and tips on shares and stocks with Kiwi trading professionals and non-professionals.
Our AI Player outperforms human professionals in the Japanese boardgame leaderboard with 40% margin. We adjusted the architecture of AlphaGo Zero for discrete huge action space.
✨ "Architected the Xilos platform — an LLM firewall designed to protect enterprises and government institutions from AI abuse and compliance risks.
✨ Defined tiered customer access levels and corresponding LLM integration layers for seamless deployment into existing platforms.
✨ Engineered a multi-layered LLM hierarchy to detect malicious queries, enrich contextual prompts, intelligently route requests and leverage RAG/CAG-based caching within a secure computation sandbox.
✨ Built the bridging infrastructure using FastAPI, LangChain, Qdrant, Pinecone, AWS SageMaker, vLLM and other modern tools.
✨ Developed deep integration with the native user interface, Ask Alpha.
✨ Led the development of the technical roadmap, defined system architecture, and outlined detailed component requirements to align with business objectives.
✨ Supported fundraising efforts by preparing technical materials and contributing to investor presentations.
✨ Designed and developed AI-integrated backend systems using FastAPI (Python), MongoDB and OpenAI API, supporting up to 1,000 concurrent accesses and serving a total of 500,000 users.
✨ Built a real-time speech recognition system with 35ms latency and 97% accuracy, setting industry-leading benchmarks.
✨ Engineered a real-time translation module that revolutionized multilingual meeting experiences, driving a 500% increase in user adoption.
✨ Created sophisticated AI models and algorithms to detect the end of participants' speech within 500ms by combining silence detection and sentiment analysis, improving key moment detection accuracy from 60% to 90% as a result.
✨ Developed an AI-powered hint generation model leveraging LLMs and Retrieval-Augmented Generation (RAG) technologies, achieving 99%+ user satisfaction.
✨ Partnered with the product team to optimize user satisfaction and market fit, contributing to scaling the user base from zero to 500,000.
✨ Designed and implemented an IoT management platform using cloud technologies and IoT protocols (MQTT, Apache Kafka), enabling the Expo City Dubai edge-tech ecosystem to support 3 million IoT devices.
✨ Developed backend systems with Django, Node.js, Java Spring Boot, PostgreSQL, Apache Airflow and AWS S3, creating seamless ETL pipelines and managing data migration with a throughput of 1TB/week.
✨ Built customized machine learning models for the Expo City Dubai IoT ecosystem, enabling anomaly detection up to 20hr in advance and achieving 95% accuracy in power consumption forecasting.
✨ Conducted research on the feasibility of adopting Chronos and other LLM-based time-series forecasting models for streaming IoT sensor metrics, leading to Expo City Dubai’s approval and catalyzing the launch of a $2.5M startup.
✨ Advanced NSFW image detection research for moderated business environments by fine-tuning CLIP-based image processing models, achieving 88% benchmark accuracy — outperforming the prior limit of 82% attained by YOLO.
✨ Investigated and implemented POC for miner prediction based on AI regression model of Bitcoin nonce distribution, with 18% of matching accuracy and $300k additional monthly revenue from it.
✨ Developed a microscope-based blood cell detection and classification system to assist clinical experts in analyzing blood samples. Utilized expertise in computer vision, object detection/classification, and mathematical modeling to achieve 98% counting accuracy, reducing testing time by 70% and cutting costs by 80%.
✨ Designed and implemented a dedicated ML inference framework optimized for low-performance, resource-constrained devices, coding in low-level C/C++ to support complex ML architectures. Delivered a 21% improvement in inference speed compared to standard public frameworks. Engineered machine learning models to predict anomalies in ECG signals from Holter devices, integrating the optimized ML inference framework to process data with less than 1-minute delay, even on low-resource wearable devices.
✨ Developed Django-based backends and contributed as a DevOps engineer for the Mega Millions lottery brokerage platform and Logi-Lab social event networking system, collectively supporting over 1 million users.
✨ Spearheaded a milestone update of Kiwi Bank’s financial service infrastructure, boosting banking efficiency by 25% and transaction speed by 10%.
✨ Designed and implemented a microservice for Stockfox financial advisory service, enabling EDGAR data scraping and insider transaction analysis using statistical modeling and Monte Carlo simulations, improving portfolio management efficiency by 150%.
✨ Managed the DevOps infrastructure for Stockfox, ensuring seamless operations across AWS, Azure and Cloudflare.
✨ Built embedded device software for football players and referees, enabling real-time audio, video and kinetic data processing through ML models.
✨ Designed outputs, including kinetics, audio feedback and AR display on dedicated glasses.
✨ Engineered an AI Stern-halma Plus Player leveraging Reinforcement Learning and Game Theory, surpassing professional human players with 40% margin in public leaderboard.
Anytime automatic algorithm selection for knapsack
In this paper, we present a new approach for Automatic Algorithm Selection. In this new procedure, we feed the predictor of the best algorithm choice with a runtime limit for the solvers. Hence, the machine learning model should consider and learn from the Anytime Behavior of the solvers, together with features characterizing each instance. For this purpose, we propose a general Framework and apply it to the Knapsack problem. Thus, we created a large and diverse dataset of instances, recorded the anytime behavior of 8 solvers on them and trained and tested three machine learning strategies, collecting the results for different machine learning algorithms. Our results show that, for the majority of the tuples instance, time, the solver that computes the best objective value can be predicted. We also make this data publicly available, as a challenge for the community to work in this problem and propose new and better machine learning models and solvers.
Don’t just take my word for it — see what clients and collaborators have to say about working with me. Their experiences reflect the dedication and passion I bring to every project.