Julian Ortega

Staff Engineer

• Shipped hybrid search combining BM25 with BGE-M3 dense retrieval across the search platform, with hot-swap rollout discipline and zero downtime to upstream systems
• Laid the foundation for an agentic merchandiser: LLMs plan, query, refine, and reason over the search and merchandising surface via tool calls
• Designed and built a working PoC of the K8s migration for a core search service of 500+ servers requiring zero downtime or upstream changes

Staff Engineer

• Evolved conversational commerce from research to production, adopting a single-agent-with-tools architecture at 20x lower LLM cost than the initial vendor; the pattern was later validated as the approach adopted by leading AI providers
• Contributed an open-source plugin PR merged upstream to dify-official-plugins, enabling the agent to reason using customer search and merchandising engines
• Built AI knowledge assistants targeting ~18% of all support ticket volume, with internal trial validating ~80% documentation coverage and a 30% ticket deflection target; the evaluation harness was its own workstream and shipped before the feature
• Led architectural redesign of the suggestion ranking system, reducing index processing times by 64-93% (Swarovski 7h42m to 31m) and improving suggestion relevance by 15-18% in customer quality testing
• Owned service lifecycle end-to-end: model deployment, real-time inference monitoring, CI/CD, evaluation metrics, and cross-functional delivery

Software Architect

• Architected and shipped AI-powered semantic search adopted by enterprise retailers across fashion, sports, grocery, and DIY verticals, supporting 40+ languages
• Led vector search scaling investigation across 4 backends, identifying the cost-viable architecture for enterprise-scale catalogues
• Built serving infrastructure handling 100x traffic spikes (5K to 500K rpm) through autoscaling architecture validated during production incidents
• Drove CI/CD optimization delivering 3x build speedup at lower infrastructure cost across the engineering organization

Team Leader / Line Manager

• Designed and shipped the Adaptive Personalization Engine (APE) from PoC to production, a microservices architecture with 6 core components processing real-time user activity streams at 200+ events/second across 4 major releases
• Delivered personalized product recommendations and search re-ranking while maintaining search response impact under 50ms average
• Initiated and led the search modernization project, reducing reindex times from 2+ hours to 26 minutes and improving search response from 192ms to 130ms; achieved 99.8% of queries under 200ms
• Mentored junior developers and facilitated career growth, with 70% of team members receiving promotions during tenure

Senior Software Engineer

• Created real-time recommendations engine proof of concept, integrating it into the core e-commerce product and establishing the foundation for production recommendation systems
• Optimized product catalog ETL processes for 32M+ product catalogues, reducing reindexing time from 40 hours to 4 hours

Software Engineer

• Developed and maintained search and recommendation algorithms processing terabytes of e-commerce data
• Implemented collaborative filtering techniques that improved recommendation accuracy across multiple client implementations

R&D Intern

• Automated and industrialized evaluation process of recommendations, creating frameworks still used in production systems
• Generated new recommendation approaches based on collaborative filtering, contributing to patent applications

Conversational Commerce Platform (Single-Agent-with-Tools)

• Self-initiated from research through PoC to production deployment, providing a narrative and live demo that supported new prospect engagement
• Evolved architecture from modular intent routing to a single-agent-with-tools pattern with templated prompts and emotional tone adaptation
• Evaluated LLM providers and reached a 20x lower cost than the initial vendor, with gains in intelligence, tool use, and speed
• PR merged upstream to dify-official-plugins exposing customer search and merchandising engines as agent tools

AI-Powered Knowledge Assistants (RAG + Tool-Calling)

• Targeted the largest addressable support category (~18% of all ticket volume), reducing resolution friction for common how-to questions
• Internal trial validated ~80% documentation coverage for real customer questions
• Designed for measurable ticket deflection with a 30% reduction target, directly reducing average resolution times of 20+ days
• Processed and embedded ~900 knowledge base pages and anonymized customer tickets into the retrieval pipeline

Hybrid Search (BM25 + BGE-M3) and Agentic Merchandiser Foundation

• Hot-swap rollout to enterprise retailers with zero changes to upstream systems
• Established the tool-calling surface for an agentic merchandiser to plan, query, refine, and reason over the catalogue

AI-Powered Semantic Search Platform

• Adopted by enterprise retailers across fashion, sports, grocery, and DIY verticals, supporting 40+ languages
• Led vector search scaling investigation across 4 backends, identifying the cost-viable architecture for enterprise-scale catalogues
• Architected dual delivery model: standalone hybrid search engine and AI Search plugin using shared embedding technology
• Handled 100x traffic spikes through autoscaling architecture validated during production incidents

Adaptive Personalization Engine (PoC to Production)

• Created the original proof of concept and evolved it to production across 4 major releases over 2.5 years
• Built microservices architecture with 6 core components processing real-time user activity streams at 200+ events/second
• Maintained search response time impact under 50ms average while delivering personalized recommendations

Master's in Computer Science (AI Specialization)

B.S. in Systems Engineering