Work History
Professional Experience
- Multi-tenant search infrastructure: Designed multi-tenant vector search service (Qdrant) with namespace-level isolation across 100+ independent environments; collection filters enforced strict data boundaries, reducing memory overhead 80% vs. per-tenant database approach.
- Hybrid search service: Built search service combining inverted-index keyword matching with semantic vector retrieval to handle exact alphanumeric codes and natural language in a unified pipeline; improved Recall@5 by up to 20% over vector-only architecture with no latency regression.
- Low-latency query routing layer: Engineered intent-classification service that routes requests before triggering expensive downstream calls; fast-path circuit breaker for direct-lookup queries reduced p99 retrieval latency from 150ms to <5ms (30x), eliminating vector search overhead for targeted file queries.
- Output validation service: Built post-generation validation microservice that fuzzy-matches LLM output against an indexed SQL tag database before rendering; achieved >99% redaction accuracy of fabricated hardware identifiers.
- Deployment infrastructure: Owned end-to-end CI/CD pipeline (GitHub Actions + Docker + AWS) for AI microservices; automated testing, container builds, and blue-green deployments enabling zero-downtime.
Python
Java
PLC Programming
Pandas
NumPy
- Large-scale text processing pipeline: Built batch embedding pipeline processing 1M+ customer records using sentence transformer models; outputs fed a segmentation service enabling precision behavioral targeting that improved marketing ROI by 20%.
- Distributed data ingestion service: Designed Apache Airflow DAG-based ETL system on AWS EC2 ingesting and transforming 3.5M+ records/month into Redshift; retry orchestration and task-level fault isolation reduced pipeline processing latency by 35%.
- Model serving & versioning platform: Built orchestration service connecting forecasting model outputs to downstream consumers; CI/CD pipelines automated model versioning, artifact promotion, and rollback, enabling zero-downtime model updates in production.
Machine Learning
PowerBI
SQL
Deep Learning
RFM Modeling
- Conducted data analysis using NLP techniques to classify and predict outcomes, extracting topics, sentiment, and named entities from large text corpora.
- Led research projects on television advertising, social media, and political marketing under Prof. Beth Fossen and Prof. Lopo Rego.
NLP
Seaborn
Feature Engineering
Machine Learning
Excel
- A/B testing infrastructure: Built traffic-routing framework across application variants; SQL-based conversion analysis drove a 17% lift in user engagement.
- Recommendation engine: Engineered collaborative + content-based filtering backend, optimizing preference retrieval queries to achieve 20% engagement increase.
- API performance: Developed Node.js RESTful APIs connecting frontends to ML microservices, cutting API response latency by 40% and driving 15% growth in daily active usage.
Statistical Modeling
A/B Testing
SQL
Recommendation Systems
- Healthcare platform: Designed microservices web platform (Java, Django, AngularJS, SQL) automating health information management; 40% reduction in query execution time through relational DB optimization.
- Cloud infrastructure: Migrated and maintained AWS EC2/RDS infrastructure, building fault-tolerant ETL pipelines with high availability for large-scale medical data storage.
- Deep learning deployment: Deployed TensorFlow/CNN diagnostic models via Docker with FastAPI endpoints, reducing inference latency by 30% and improving system uptime.
TensorFlow
AWS EC2 / RDS
CNN
Node.js
ETL
- Researched incremental/decremental versions of DBSCAN and MBSCAN algorithms for distance-based outlier mining.
- Used Python and C++ to improve computational speed by 30% and precision by 20% in clustering algorithms, tested on a fintech application with over 1 million users for customer segmentation and portfolio optimization.
Python
C++
Clustering
Data Analysis
Machine Learning
