Vikram Tharakan

I’m a Machine Learning Engineer who builds production-ready intelligence systems that turn messy, heterogeneous data into actionable insights—often using graph models, NLP, and real-time entity resolution.

At Booz Allen Hamilton, I architected and led delivery of a large-scale data fusion platform that ingests vendor datasets, normalizes content via Apache Tika, and transforms it into a unified graph data model with Databricks, Apache Spark, NiFi, AWS, Elasticsearch, and JanusGraph. Within a ~10-person team, I drove end-to-end delivery under tight deadlines—spanning data engineering, integration, and deployment—so the resulting service is actively used by agents in the field.

I’ve also focused on high-impact data intelligence pipelines: I contributed to an entity resolution workflow combining Fellegi-Sunter probabilistic linkage with rule-based candidate generation, and I engineered a scalable entity deletion service from the ground up with cascading dependency resolution across multi-source linked graphs.

Earlier, at MITRE, I developed ETL microservices for document ingestion into Elasticsearch using NiFi, Tika, OCR, caching, and neural machine translation. I fine-tuned extractive QA transformer models (NER/EQA) with an active learning loop to reach strong F1 performance, and I co-developed a deep metric learning recommender system over geospatial data—then deployed relationship graphs in Neo4j to surface cross-source linkages.