I take AI from a first idea to a system that holds up in production — and I own the gap in between, where most projects quietly die. The value sits at both ends of that arc: a POC run as an honest experiment that de-risks the six-figure bet before you place it, and the engineering to keep the winner correct under failure, secure by design, and cost-bounded at scale. I’ve built this across Fintech, Healthcare, Legal Tech, Renewable Energy, civil engineering, and sports media — one repeated problem shape, not a scatter of verticals — spanning LLM pipelines, classical ML, and the data infrastructure beneath them.

• Architected an incremental map-reduce LLM pipeline that stays correct under a distributed retry engine — idempotent under at-least-once execution, fail-closed on partial coverage, with per-run LLM cost proportional to new activity rather than total history, and prompt-injection defense on every model output that reaches persisted state
• Built a production multi-stage extraction pipeline — parallel extraction → independent judge verification → curation — that produces factually grounded output from noisy speech-to-text, after a structured four-vendor LLM evaluation (Claude, OpenAI, Gemini, Grok) and prompt-cache cost optimization
• Engineered the production platform for a public-facing conversational AI at one of the largest privately held US companies — Pulumi infrastructure-as-code across three AWS environments, GitOps delivery, in-cluster LLM observability with no data egress, and unified multi-provider authentication
• Led an interdisciplinary team as technical lead and architect building AI document automation that cut report compilation from days to hours, with LangGraph agents, structured output, and human-in-the-loop verification in a client-facing role
• Built the distributed data infrastructure that makes the AI reliable in the first place — event-driven blockchain pipelines on Kafka and bulk ETL 10x faster than the state of the art (8 hours vs. weeks; 400GB–1TB Neo4j graphs) — because trustworthy AI output is an ingestion, orchestration, and validation problem before it is a model problem

Architected multi-tenant data platform and AI-powered financial analyst for cryptocurrency products, combining event-driven and batch data ingestion with defense-in-depth security.

• Built unified data platform with Kafka Connect event-driven replication and Dagster-orchestrated batch API ingestion into PostgreSQL RDS
• Implemented automated Row-Level Security via DBT macros ensuring business-level data isolation across all analytical models
• Designed AI-powered natural language financial analyst using LangGraph and AWS Bedrock Claude with 5-layer security architecture reducing query time from hours to seconds while achieving zero data breaches across multi-tenant deployment
• Deployed self-service analytics infrastructure using Apache Superset and Chainlit chat interface for business stakeholders

Led data architecture and analytics initiatives for cryptocurrency products, combining on-chain and off-chain data to deliver scalable, production-grade features.

• Architected and deployed data-driven features leveraging blockchain and external data sources
• Built robust data pipelines for ingesting and transforming blockchain data at scale
• Developed real-time analytics infrastructure using Python, Docker, Dagster, Kubernetes, Kafka, Superset, MongoDB, Neo4j, Postgres, Grafana, and AWS
• Led cross-functional team (DS, DE, BI, engineers) while remaining fully hands-on in system architecture and implementation

Developed production-grade systems for analytics, predictive modeling, and fraud detection.

Led AI strategy and infrastructure development for a data-driven product intelligence platform.

• Defined the AI roadmap and built foundational infrastructure for long-term product intelligence
• Introduced automated model evaluation frameworks to support experimentation and deployment
• Collaborated with engineering and product teams to align AI capabilities with business needs

Designed data ingestion and processing flows integrating external service providers.

Built end-to-end ETL pipelines using Apache NiFi and Python. Trained ML models (TensorFlow, XGBoost) for compound analysis and prediction tasks. Designed modular, reproducible workflows for model training and experimentation.

Defined AI roadmap for non-invasive glucose monitoring. Hands-on work in device calibration, noise reduction, and personalized medicine integration for AI-based glucose monitoring.

Led AI and data science initiatives for medical diagnostics and blood management solutions. Built ML models and expert systems (Python, TensorFlow, XGBoost) with explainable AI using clinical decision rules.

Led data science and ML for an analytics platform. Built anomaly detection engine for BI metrics using InfluxDB, Python, and NodeJS. Managed end-to-end ML pipeline from data acquisition to deployment on AWS. Architected scalable backend infrastructure using microservices.

Conducted advanced research in distributed dynamic graph data processing and Linked Data systems while leading teaching and research coordination activities.

• Designed and implemented a provenance-aware system for efficient processing of static and dynamic graph data (Linked Data)
• Enhanced system capabilities to support hypothetical queries and advanced data analysis
• Developed data processing systems using Linux, C/C++, and Python
• Developed, authored, and taught three courses:
  • Semantic Web and Linked Data (lectures and labs),
  • Semantic Web and Linked Data (research seminar),
  • NoSQL databases (lectures and labs)
• Supervised MSc and PhD students on their thesis projects
• Actively contributed to grant acquisition for new research projects
• Coordinated research efforts and managed collaborative projects with scientific teams
• Co-authored three books on Big Data:
  • Handbook of Big Data Technologies
  • Linked Data: Storing, Querying, and Reasoning
  • Encyclopedia of Big Data Technologies

Conducted research and development in Semantic Web and Big Data technologies at one of Europe’s leading applied research institutes.

• Led workshops and training sessions on Semantic Web and Linked Data technologies
• Actively contributed to grant proposal preparation and project acquisition

Conducted research in the eXascale Infolab under Professor Philippe Cudré-Mauroux, focusing on distributed database systems for Linked Data.

• Designed and developed an efficient, scalable, and provenance-aware distributed database engine for Linked Data processing
• Led evaluation project for NoSQL databases in Linked Data applications, including benchmark scenario design
• Built systems using Linux, C/C++, AWS, NoSQL databases, and Python for large-scale data processing
• Taught courses on MapReduce and Linked Data technologies
• Supervised graduate students on research projects and thesis work

Built foundational expertise in database systems, Linux infrastructure, and production systems across enterprise and startup environments. Managed critical infrastructure, designed data architectures, and delivered reliable systems under diverse technical constraints—establishing the systems thinking and operational discipline that underpins current data and AI work.