Dmitry Kazhdan, PhD

Working in the FinCrime Department, applying latest advancements in AI / LLMs / Agentic Workflows to power the next generation of financial products.

General: Worked closely with the CEO to shape the strategic vision of Tenyks. Contributed to raising over $3.4M from top-tier VCs by shaping and pitching the technical vision to investors. Product: Oversaw ideation, design, architecture, research prototyping, & delivery of the two Tenyks products: an MLOps platform for Computer Vision model debugging, and an AI-first surveillance platform for camera stream analytics. The latter product processed 100s of camera streams in realtime via novel model cascade architectures combining multi-modal LLMs (e.g. Gemini) with powerful object detectors & VLMs, successfully detecting crucial events such as patient accidents in care homes, or operational inefficiencies in Quick-Service Restaurants. Hiring: together with the Founding team, grew Tenyks to 15 FTEs (at peak). Designed and ran the technical hiring pipeline (incl. take-home assessments, interviews, followups). Led a technical team of 9 FTEs (across SW, ML, Design, and Support). Engineering: collaborated with Tenyks engineers in architecting the Tenyks products across the stack including: FE (React), BE (FastAPI/Flask), DBs (Mongo/Timescale), Infra (K8s), Cloud (AWS/GCP). Led the delivery efforts of SOC-II certification (using the Vanta platform). Awards: YCombinator S21 (2021); Cambridge Computer Lab Company of the Year Award (2022); Forbes 30U30 Europe, Technology (2023); Embedded Vision Summit Edge AI and Vision Product of the Year Award (2024);

My role at CUE involved finding and securing investors that share CUE’s vision, in order to raise funds for fuelling the wide range of CUE’s initiatives and events.

Worked on Continual Learning approaches to personalised Machine Learning systems leveraging sparse, user-generated data. Built dynamic Machine Learning systems capable of adapting to individual users & usage patterns, providing a quality personalised service at a lower deployment cost.

Worked in the Arm Compute Library team, building an edge Face Recognition with Face Detection pipeline for Arm-based platforms using Arm Compute Library functionality.

Worked in the Toolbelt team implementing new features of the Redgate Product Installer.

Title: "Enhancing Interpretability: The Role of Concept-based Explanations Across Data Types" Focusing on novel approaches to Concept-based Explanations for Deep Neural Networks across multiple modalities (including Computer Vision, Temporal/Tabular, and Graph).

Thesis: MARLeME - a Multi-Agent Reinforcement Learning Model Extraction Library. Published at the International Joint Conference on Neural Networks (IJCNN).