IEEE
Published Research
WMSM on real-time Hebrew Sign Language recognition and sentence-level translation.
Israel-based Lead ML/AI Engineer | English-language delivery across Europe and international teams | IEEE-published researcher | MBA candidate
Research-backed AI systems for Europe and international teams.
אייל פשה · מפתח וחוקר AI/ML · ישראל
Based in Israel, I work in English across Europe and international teams on enterprise AI systems, Hebrew NLP, and production machine learning products that need to hold up under real operational conditions.
I architect and lead enterprise AI systems from Israel while collaborating in English with Europe and international teams. The core work spans full-stack infrastructure, custom model training, multi-model orchestration across OpenAI, Gemini, and local LLMs via vLLM, and production-grade NLP for organizations that need usable systems rather than isolated demos.
My research produced WMSM, a real-time Hebrew Sign Language recognition and sentence-level translation framework published by IEEE. It reached 97.29% word-level accuracy and 21.15 BLEU-4 while cutting training costs by 99.93%, giving the work a public academic record that is legible well beyond the local Israeli market.
Beyond technical leadership, I bring almost a decade of entrepreneurial experience as the founder and manager of Jerseys Club, an e-commerce venture I scaled to six-figure annual revenue and more than 10,000 customers. That founder mindset helps me work effectively with distributed stakeholders, commercial constraints, and cross-functional teams across regions.
MBA
Interdisciplinary Thesis
Computer Science and Business Administration focused on deep learning commercialization.
Multi-model
AI Orchestration
Production workflows spanning OpenAI, Gemini, and local LLMs via vLLM.
10k+
Customers Served
Founder-led commerce platform scaled to six-figure annual revenue.
About
Research depth, product ownership, and commercial context for teams beyond one market.
The work is rooted in Israel but structured to be useful for Europe and international teams evaluating technical depth, delivery ownership, and commercial judgment.
The through-line is building strong technical work into systems teams across regions can actually use, fund, and operate.
That is why the MBA is not presented here as a generic business credential. The thesis is interdisciplinary by design, focused on the intersection of deep learning and business administration.
It pairs with a B.Sc. in Computer Science focused on machine learning, deep learning, and real-time systems, and with work that spans model architecture, data engineering, backend systems, frontend delivery, and product ownership.
Whether the problem is clause risk detection, real-time sign language translation, or operational automation, I tend to work from the full stack outward until the system is usable, reliable, and clearly valuable in practice for local and cross-border stakeholders alike.
Core Stack
LLM OrchestrationNLP PipelinesComputer VisionTransformersPythonTypeScriptPyTorchTensorFlowNode.jsPostgreSQLReactNext.jsAzureAWS
Experience
Roles with measurable outcomes across enterprise AI, research, and growth.
The recent work ranges from production LegalTech systems and academic research to founder-led commerce and frontend design.
November 2025 - Present
Lead ML/AI Engineer
Sensemaking Israel Rehovot, Israel
Architecting and leading the end-to-end development of the first AI-native operating system for law firms, an enterprise LegalTech SaaS platform, owning everything from full-stack infrastructure to production ML deployment.
- Engineered a multi-model AI stack utilizing OpenAI, Gemini, and local LLMs via vLLM, building custom agents to flag high-risk indemnity clauses and extract structured legal data.
- Designed a highly scalable, Azure-hosted PostgreSQL architecture to support high-availability document processing for demanding enterprise workloads.
- Collaborated closely with Microsoft product teams to drive deep integrations across Microsoft 365 and Microsoft Graph, keeping the product inside the workflows legal professionals already use.
- Built and deployed production-grade NLP pipelines using Python and TensorFlow for document classification, information extraction, and predictive case analytics.
January 2024 - September 2025
ML Research Developer
The College of Management Academic Studies Rishon LeZion, Israel
Led a two-year research initiative to develop WMSM, a real-time Hebrew Sign Language (HSL) framework, culminating in a formal IEEE publication.
- Engineered a three-phase augmentation pipeline that expanded 2,342 seed videos into 252,712 robust training samples.
- Achieved 97.29% word-level accuracy, 99.72% Top-3 accuracy, and a 21.15 BLEU-4 score, significantly outperforming Seq2Seq and transformer-based baselines.
- Optimized compute efficiency, reducing training time to just 24 GPU-hours - a 99.93% reduction compared to existing industry benchmarks.
- Translated the research into a live product by integrating the model into Handibur, an iOS video chat beta delivering HSL-to-text at 40 FPS and securing 2nd place in a collegiate Shark Tank competition.
November 2018 - January 2025
Founder and Business Owner
Jerseys Club Israel
Founded and scaled an e-commerce business for over 7 years from zero to six-figure annual revenue and more than 10,000 customers while owning the full technical roadmap.
- Built and maintained a high-volume platform with React and Angular frontends supported by a custom backend architecture.
- Implemented Celery and asynchronous workflows to automate order fulfillment and reduce operational overhead.
- Introduced production ML components for customer data optimization and supply chain logistics.
- Led outsourced development teams with Jira and drove growth through Google Analytics, Search Console, and Meta Ads.
May 2020 - April 2021
Frontend Developer
H.S. Diamonds Ltd Ramat Gan, Israel
Led the redesign and performance overhaul of a luxury jewelry storefront, aligning premium UX with performance and conversion improvements.
- Redesigned the responsive storefront with a mobile-first and performance-first approach that improved the shopping experience across devices.
- Improved the site's structure and technical foundation to support product discovery and buyer confidence.
Selected work
Three builds that explain the range.
Private enterprise AI, published research that later became a product, and a founder-built commercial platform.
Private production platform
Enterprise LegalTech platform
An AI-native operating system for law firms, combining clause intelligence, complex legal and medical file analysis, case visibility, and communication inside one production system.
- Multi-model orchestration across OpenAI, Gemini, and local LLMs via vLLM to power custom legal-document agents.
- Scalable Azure PostgreSQL foundation architected to manage high-volume enterprise data and rigorous legal workflows.
- Deep Microsoft 365 integration - developed in close collaboration with Microsoft product teams to seamlessly embed the platform inside our system.
IEEE publication to live beta
WMSM + Handibur
A lightweight framework for real-time Hebrew Sign Language recognition and sentence assembly, later published by IEEE and integrated into a live iOS video chat experience.
- 97.29% word-level accuracy, 99.72% Top-3 accuracy, and 21.15 BLEU-4.
- 24 GPU-hours total training, reducing compute cost by 99.93% versus comparable approaches.
- Live HSL-to-text delivery in Handibur at 40 FPS.
- Presented internationally in Vienna and Israel before moving from research into a product-facing mobile beta.
Founder-built business infrastructure
Jerseys Club commerce engine
A full-stack commerce platform that evolved from manual operations into an automated, data-informed system with end-to-end ownership across engineering, operations, and commercial decision-making.
- Scaled to six-figure annual revenue and more than 10,000 customers.
- Automated fulfillment and operations with Celery-driven asynchronous workflows.
- Combined product engineering, growth analytics, financial planning, and supply chain execution.
- Built the internal backbone across storefront flow, supplier coordination, reporting, and customer-facing operations.
Research
WMSM: efficient real-time Hebrew Sign Language recognition.
Lead-author IEEE publication on a lightweight framework for word recognition and sentence assembly, later integrated into a live iOS beta.
Publication
WMSM: An Efficient Real-Time Framework for Hebrew Sign Language Recognition and Sentence-Level Translation
Lead-author publication in IEEE FLLM 2025 on an efficient framework for real-time Hebrew Sign Language recognition and sentence-level translation.
WMSM replaces heavy sequence-to-sequence translation with an efficient word-level model and a sentence mechanism designed for real-time Hebrew Sign Language recognition. The architecture shows that x-y hand landmarks can be enough for accurate gesture recognition, while a custom loss function and temporal sentence assembly process keep predictions stable in live video.
- Reached 97.29% word-level accuracy, 99.72% Top-3 accuracy, and 21.15 BLEU-4.
- Reduced training to 24 GPU-hours, a 99.93% drop in computational cost compared with existing benchmarks.
- Expanded 2,342 videos into 252,712 training samples through a three-stage augmentation pipeline.
- Presented internationally in Vienna and Israel, then integrated into the Handibur iOS beta.
GitHub
Recent public repositories from GitHub.
10 public repositories currently on GitHub.
TypeScript + Python
BundesTrip
Plan sports trips like a pro with multi-city travel, optimized train routes, hotel stays, and football-packed day-to-day itineraries across Germany. Live product — source is private.
Python
smart-extract
GPU-accelerated tool that scans presentation videos, scores visual quality, and extracts the best frames using computer vision and AI upscaling.
TypeScript
Redzone-Companion-My-Players
A professional fantasy football tracking application that allows you to monitor your players across multiple Sleeper leagues during live NFL games. Perfect for RedZone viewing sessions!
Python
LLM-Travel-Agent
A sophisticated conversational travel assistant powered by advanced prompt engineering and psychological profiling.
Jupyter Notebook
Pneumonia-Detection
Deep learning models (CNN, VGG-16, DenseNet) for detecting pneumonia from chest X-rays. Includes training, evaluation, t-SNE visualizations, and more.
Jupyter Notebook
ML-Fashion-MNIST-Classification
Binary clothing classifier using Fashion MNIST. Built with TensorFlow/Keras, includes preprocessing, training, and evaluation with 81%+ accuracy.
Posts
Conference updates and publication notes from LinkedIn.
Recent LinkedIn posts covering research milestones and public updates.
Recent LinkedIn update
IEEE FLLM 2025 publication update
Contact
AI architecture, applied ML, and research collaboration for Europe and international teams.
Open to conversations with Europe and international teams about enterprise AI systems, deep learning commercialization, research collaboration, and full-stack product delivery.
Email or LinkedIn is the fastest way to start.
I am based in Israel and work in English across research, production ML systems, and product architecture. If the work involves moving a technical idea into something deployable, commercially usable, and clear to cross-functional stakeholders across borders, it is likely relevant.
2025 - 2026
MBA with interdisciplinary thesis
Computer Science and Business Administration, The College of Management Academic Studies
2021 - 2024
B.Sc. in Computer Science
Focus on machine learning, deep learning, and real-time systems, The College of Management Academic Studies
Explore
More to read.
A few deeper reads on the AI/ML work, the LegalTech platform, the academic background, and the research.
Explore
More to read.
A few deeper reads on the AI/ML work, the LegalTech platform, the academic background, and the research.
International
Israel-based AI/ML engineer for Europe and international teams.
English-language AI architecture, Hebrew NLP, and research-backed delivery for Europe and international organizations.
Read moreIsrael
AI/ML engineer in Israel focused on Hebrew NLP and applied ML systems.
A closer look at Israel-based AI/ML work across Hebrew AI, applied machine learning, and production systems.
Read moreLegalTech AI
Hebrew NLP and LegalTech AI in production legal workflows.
A closer look at enterprise document intelligence, multi-model orchestration, and Hebrew-English AI delivery for cross-border legal workflows.
Read moreResearch
WMSM research on Hebrew Sign Language recognition.
A closer look at the WMSM publication, the key results, and the international research-to-product path into Handibur.
Read moreAcademic
Academic AI research in Israel, from IEEE publication to interdisciplinary thesis.
A closer look at the publication record, international conference footprint, COLMAN background, and the interdisciplinary thesis.
Read moreEyal Pasha (אייל פשה) is a Lead ML/AI Engineer and IEEE-published researcher based in Israel, building the first AI-native operating system for law firms at Sensemaking Israel and lead author of the WMSM Hebrew Sign Language recognition framework.