About

Work Experience

• ML Platform Engineer, Pipedrive

  May, 2025 - Present

  Contextual intelligence platform for agentic AI workflows, plus ML platform infrastructure.

  • Designed and scaled platform capabilities for agentic AI workflows, cutting delivery of new agentic features from weeks to hours.

  • Shaped the system design of a graph-native memory and vector retrieval layer, taking it from hypothesis and proof of concept to a scalable service processing millions of records per day.

  • Replaced duplicated per-team solutions with shared infrastructure, shortening development time for at least three teams and improving output quality and provenance through traceable, artifact-based memory.

  • Led a cross-team initiative to an alpha release of an internal LLM copilot.

  • Migrated multiple production analytics services on Kubernetes with Helm, including Airflow serving many data projects and users, without major disruption; provisioned and managed AWS resources with Terraform.

• Machine Learning Engineer, Paxful / Noones

  Jun, 2022 - Apr, 20252 years 10 months

  Owned the end-to-end MLOps lifecycle for fraud prevention, security and customer support automation on a crypto marketplace: ETL pipelines, feature engineering, model training and deployment, monitoring, error analysis, LLM integration, and intern supervision.

• Machine Learning Engineer, Quantum

  Aug, 2019 - Aug, 20212 years

  Led multiple production ML projects spanning geospatial analysis (crop classification, segmentation on Sentinel-2/Planet imagery), agricultural computer vision (disease detection, flower segmentation), document intelligence (OCR, classification, NLP parsing), and mobile ML (Apple CreateML/CoreML for sports analytics). Drove the full ML lifecycle from research and model development through deployment and team coordination.

• Machine Learning Intern, Quantum

  Apr, 2019 - Aug, 20194 months

  Developed a forest clear-cut segmentation system for Sentinel-2 satellite imagery using deep learning models (U-Net, FPN+ResNet). Built the complete solution, including a GIS processing pipeline and a web service backend with Django, Docker and PostGIS.

• Machine Learning Intern, CHI Software

  Mar, 2019 - Apr, 20191 month

  Worked with classical ML models (linear/logistic regression, clustering, SVM, bagging, boosting), deep learning, and time series.

Projects Experience

• Project Tapestry (AI Alliance)

  - Present

  Contributor to Project Tapestry, an open consortium effort to train a shared frontier model in which partners keep ownership of their data and compute. Contributed formal specifications in Quint of the consortium's governance and its federated training protocol (how participants submit, verify and aggregate model updates), testing both for capture risks and protocol-level failures. Currently developing the project's threat model.

• Grey / Alley OS

  - Present

  Grey was an open-source Rust implementation of the JAM (Join-Accumulate Machine) protocol client, the next-generation Polkadot architecture. The project has since grown into Alley OS, a monorepo for JAR, a JAM-derived chain protocol with a Lean 4 formal specification and a Rust implementation. Contributed the GRANDPA finality component to the Grey client.

• Decentralized Relay

  - Present

  Fork of Relay, a real-time collaborative editing plugin for Obsidian, that removes the central server. Note updates sync peer-to-peer over WebRTC, so document content never leaves users' devices; optional encrypted snapshots on Polkadot Bulletin Chain let peers recover when no one else is online. Identity uses hardware-backed passkeys with per-device proxy keys on Westend Asset Hub, and folder access is enforced on-chain through role-aware invite codes, with no custom pallet or relay server.

• fl-chain-data-sharing

  - Present

  Proof of concept of a blockchain-based federated learning system with zero-knowledge proofs for verifying model aggregation. MSc thesis, University of Tartu.

Skills

Education

Awards

Volunteer Work

Publications