Ahmed – Python, AI agent development, LLM

Built a production-grade multi-agent AI system that converts natural language into data insights, dashboards, and reports by interpreting complex schemas, generating and validating SQL queries, and automatically producing visualizations and summaries.

• Took product from prototype to full production system;
• Designed end-to-end multi-agent architecture (querying, validation, visualization, insights);
• Built a semantic layer enabling accurate SQL generation across complex schemas;
• Implemented RAG pipelines for large-table retrieval and context enrichment;
• Reduced latency from 12s to 1.5s and cut retries by ~50%;
• Developed pipelines to convert unstructured Excel data into structured datasets;
• Built evaluation and validation frameworks to ensure AI output correctness;
• Led system architecture, task decomposition, and cross-functional execution;
• Collaborated directly with customers to translate ambiguous needs into AI solutions.

Improved large-scale identity resolution by designing more effective blocking strategies that increased matching accuracy while maintaining performance at scale.

• Designed advanced blocking strategies improving match coverage;
• Built an evaluation & experimentation framework for algorithm quality;
• Increased recall and matching accuracy in identity resolution;
• Enabled systematic testing of algorithmic improvements;
• Contributed to roadmap decisions around measurable ML quality improvements.

Designed and deployed infrastructure optimization systems to reduce cloud costs and improve performance across large-scale production services.

• Led company-wide DynamoDB best practices and architecture;
• Built a custom autoscaling system outperforming AWS defaults (~4 minutes faster);
• Saved ~$1.5M/year through DynamoDB architecture optimizations;
• Generated $80K+/month savings via autoscaling improvements;
• Migrated massive datasets (billions of records) safely and efficiently
• Conducted infra audits, uncovering $100K/month waste.

Designed and built NLP systems and research prototypes to improve readability, language understanding, and large-scale text analysis, resulting in multiple publications in top NLP venues.

• Built Smart Reader, a system providing contextual linguistic insights (syntax, semantics, lexical info);
• Developed algorithms for improving text readability and word wrapping;
• Conducted large-scale analysis of Arabic dialects from social media data;
• Published multiple papers in top-tier NLP conferences;
• Translated theoretical ideas into working systems and prototypes.