AI/ML leader with 20+ years building intelligent systems — from co-founding a sound-recognition startup (acquired by Analog Devices) to designing agentic AI workflows and publishing 200+ open-source Python packages. Ph.D. in Mathematics, 6 issued patents, 15+ peer-reviewed publications.
Since the generative AI inflection of 2023, my work has centered on the question most organizations face now: how do you actually wire LLMs into real workflows? I build the connective tissue — agentic pipelines, retrieval-augmented generation systems, semantic navigation tools, and no-code AI application frameworks — that turn foundation models into operational leverage.
At Figiri, I design visualization tools for graph and embedding-space exploration — enabling analysts to see, navigate, and interrogate the semantic spaces that LLMs and RAG systems depend on. At Addix Group, I helped enterprises adopt LLMs through proof-of-concept systems including resume-matching engines, contextual assistants, and RAG-powered workflows.
Artificial Intelligence eXtensions — composable tools for AI workflows.
Visualize large-scale network graphs and ML embeddings interactively.
Clean Python interface to OpenAI APIs. Composable, minimal, dict-like patterns.
Extract code exercises from code itself — structured skill definitions.
Generate MCP servers from Python functions — the py2X pattern for Model Context Protocol.
Operational stewardship tools for AI systems.
Context management for AI applications.
Backend framework for exploring and comparing RAG operations — retrievers, vector stores, and answering strategies.
Tools for working with embeddings — generation, storage, search, and analysis.
I co-founded OtoSense — a signal ML company that taught computers to "hear" sounds and "feel" vibrations. We developed language-model-inspired approaches to acoustic signals: decomposing sound into primitives, building syntactic representations, and mapping them to semantic labels. The technology served industrial monitoring, home security, accessibility for deaf users, and more.
OtoSense was acquired by Analog Devices in 2018, less than five years after inception. I then served as Director of Machine Learning at Analog Devices (2018–2023), where I built ML acceleration platforms, open-source tooling for data access and experimentation, and mentored data science teams across the company.
Building visualization and ML tools for graph and embedding-space exploration. Interactive UIs for semantic navigation in RAG, recommendation, and similarity search workflows.
Helped organizations adopt LLMs and generative AI. Delivered PoCs: resume-matching systems, no-code AI platforms, RAG-powered contextual assistants.
Led ML platforms and tooling strategy for IoT and signal ML. Built Python frameworks unifying data access, feature engineering, and experimentation. Mentored data science teams and shaped company-wide AI strategy.
Co-founded a sound-recognition AI company. Developed semantic audio models, led architecture and ML. Acquired by Analog Devices.
Attribution models, behavioral targeting, clickstream analytics, and KPI time-series for marketing intelligence.
NLP and semantic systems for AdWords optimization. Pricing, forecasting, and geographic data quality for large-scale marketing operations.
Independent ML consultant across Europe and the US. Recommender systems, Bayesian inference, duplicate detection, semantic analysis, forecasting, and optimization. Clients included Sanoma, Easyvoyage, Ricoh, Metron, and others.
A sustained, decade-long commitment to shipping reusable Python tooling. The ecosystem spans data access layers, ML pipelines, audio processing, visualization, storage abstraction, and developer utilities. Many packages are actively maintained and used in production workflows. Browse all on PyPI →
Base tools for building and transforming Data Object Layers — dict-like interfaces to any storage backend.
Link functions into callable DAG objects — the backbone of pipeline composition.
Quick HTTP web-service construction from Python functions.
Business analytics utilities.
A data-object-layer to get tables from a variety of sources with ease.
Data Object Layer for HuggingFace — dict-like access to models and datasets.
JSON schema utilities — validation, transformation, and generation.
Investigations in financial sentiment analysis.
Higher-order function utilities.
Peer-reviewed research spanning two domains: pure mathematics (graph theory, combinatorics, Hamiltonian systems) and applied social science (drug-use trajectories, longitudinal data measurement, harm reduction).
Graph theory & combinatorics. Advisor: Ronald J. Gould.
I write about Python architecture, ML tooling, and the craft of building systems that last. I also have a history of teaching probability, statistics, data science, and software design at universities and in industry workshops.