Moustafa Zein

TEAM LEADERSHIP

• Lead engineering team on LLM systems; set standards for prompt engineering, fine-tuning, and model evaluation
• Review designs/PRs; unblock execution
• Partner with domain experts (lawyers) to translate legal requirements

MULTI-AGENT ORCHESTRATION

• Designed serial/parallel workflows using Pydantic AI, LangGraph, and MCP
• 4-agent system: contract parsing → risk analysis → compliance checking → recommendations
• Lawyers review contracts 10x faster (20 hrs → 2 hr)

INFERENCE OPTIMIZATION

• Optimized Azure OpenAI inference: <100ms latency (was 30-45s)
• Prompt caching: -65% token usage, -$2K/month cost
• Constrained JSON outputs: 73% token reduction

LLM FINE-TUNING & EVALUATION

• Built an evaluation framework comparing LLM outputs vs human lawyer benchmarks
• Achieved 98% accuracy on contract risk identification
• Implemented the RLHF pipeline with lawyer feedback for domain-specific accuracy
• Integrated Pinecone RAG: legal precedent retrieval (95%+ accuracy)

LEGAL PRODUCTS SHIPPED

• Document Drafting: LLM generates contracts from templates; 2x lawyer productivity
• Contract Review: Multi-agent system flags risks, missing clauses, compliance gaps
• Precedent Search: RAG-powered retrieval of relevant case law and regulations
• Impact: Enabled legal professionals to 2x daily document throughput

TECHNICAL DECISIONS

• Choose Pydantic AI for type-safe structured outputs (critical for legal accuracy)
• LangGraph for state management (legal reasoning requires a multi-step context)
• Azure OpenAI for enterprise compliance and data residency requirements

• Hired and established the ML algorithms and data infrastructure team, and developed AI/data infrastructure using Azure cloud solutions. 
• Developed ML and LLM models to enhance user experience, particularly by reducing delivery estimated time by 20 minutes and personalizing dish recommendations. 
• Built robust data streaming and warehouse pipelines to handle 120M records daily and store 4.5TB of data over the last six months.
• Implemented an automated vehicle routing system to cover 24 areas, resulting in a 100% increase in daily orders from 9.8K to 18K within 8 months.
• Delivered both the business and technical roadmaps for ML usage and data engineering in food delivery, contributing to securing the next funding round.

• Managed a team of 15 engineers, improving coordination with product managers and business partners.
• Set up and led a machine learning engineering team to deliver key features: Autos Copilot based on LLM, My Garage, Good Offers prediction, and Price Insights model.
• Developed video and image analytics model for autos marketplace based GPU optimization, Azure distributed systems. These models generated 360 view of images and defined the videos and car junk detection
• Contributed to the international expansion of the auto marketplace product across seven countries in Europe and Asia, resulting in a 200% increase in daily active users.
• Led development of a scalable, reliable data ingestion platform for the auto marketplace, handling challenges like stale listings and duplication, and processing 20 million automotive records daily.
• Developed an LLM-powered solution that generates 2,000 shopping buying guide articles daily, boosting daily active users from 40K to 180K. I have trained OPenAi models to get more reliable and accurate results
• Established processes to stream and ingest 100K coupons, cutting partnership costs by 47%.
• Built an end-to-end integration with the Edge browser, achieving 950K daily active users within the first three months.