Experience
8+ years building ML systems at scale — from semantic search to recommendation engines.
Senior ML Engineer
Semantic Search System
Led end-to-end development of distributed, multi-tenant, multilingual semantic search system serving 10M requests/day, achieved p99 < 35ms latency and processing over 33TB event data.
- Solved critical market expansion blocker: designed language-agnostic embedding strategy enabling same-day market launch without retraining. Unlocked 5 new international markets (Germany, Netherlands, Japan, Spain)
- Identified that 73% of customer complaints stemmed from empty search results. Prioritized semantic understanding, reducing support tickets by 2,400/month ($180K annual savings)
- Improved CTR by 50%, conversion rate by 20%, reduced 67% empty search results rate — improving app store rating to 4.8 stars
- Handling multi-tenancy with data isolation, maintaining cache efficiently using consistent hashing across distributed Qdrant database cluster
- Solved multi-lingual cold-start in non-English, niche categories through curated LLM query validation at scale
- Architected distributed search infrastructure across 3 regions with active-active replication, handling 200 QPS peak traffic. Designed sharding strategy partitioning 50M embeddings by tenant, enabling horizontal scaling while maintaining 35ms p99 latency
AI Review Summarization
Built GenAI system processing 100K+ reviews for 300+ stores.
- Developed cost-optimized LLM pipeline maintaining 90% accuracy at 10x lower cost than GPT-4
- Improved merchant NPS by 4% through actionable review insights
- Implemented Map-Reduce paradigm for distributed processing with sentiment analysis
Text2SQL MCP Integration
Architected AI-powered data insights platform.
- Built authenticated MCP servers with semantic caching reducing query time by 60%
- Integrated with data warehouse enabling natural language analytics for non-technical users
ML Leadership
Established ML architecture review board and drove company-wide engineering standards.
- Established ML architecture review board, creating design patterns adopted by 6 teams
- Led RFC process for company-wide feature store implementation, reducing duplicate effort across teams by 40%
Expert Data Scientist
One Mount Group
Built AutoML platform reducing ML deployment time from weeks to days, serving 10M+ users
AutoML Recommendation Platform
Led development of multi-tenant recommendation system.
- Processed 200 GB-scale data with distributed GPU training using NVIDIA RAPIDS
- Onboarded 6 use cases, achieving 20% CTR increase and 14% conversion uplift
- Built ML pipeline with automated A/B testing, model versioning, and drift detection
Customer 360 Platform
Architected unified customer data platform for 10M+ users.
- Consolidated 200+ attributes from multiple touchpoints across fintech, e-commerce, real estate
- Led cross-functional team of 12 (Data Engineering, Governance, Science, Analytics)
- Enabled advanced segmentation driving $5M in targeted campaign revenue
Demand Forecasting
Built ML models for B2B inventory optimization.
- Reduced stockouts by 30% through time-series forecasting with external data integration
- Implemented seasonal decomposition and trend analysis for 1000+ SKUs
Data Scientist
Open Commerce Group
Built recommendation engine and data platform serving 100K+ merchants
Graph-based Recommendation Engine
Developed real-time recommendation system.
- Implemented in-memory graph traversal serving 1M+ requests/day with sub-second latency
- Built Lambda architecture with Spark, Kafka, Redis for real-time feature engineering
E-commerce Intelligence Platform
Led team building competitive intelligence tools.
- Processed 1TB+ daily logs from multiple marketplaces (Shopify, AliExpress, Taobao)
- Built analytics stack: S3, Kinesis, Lambda, Athena serving 50+ data analysts
Data Scientist
Apvera
IoT Security Startup
IoT Anomaly Detection
Developed anomaly detection for IoT security processing 100M+ events/day.
- Built Lambda architecture with Spark, Kafka, Cassandra for distributed stream processing
- Processed 100M+ events/day for real-time threat detection