Hybrid AI, knowledge graphs, and decision systems

Steve Bertolani

I build & architect AI systems that connect LLMs, knowledge graphs, predictive models, optimization, and enterprise data platforms. My work spans scientific computing, staff-level data science, customer-facing architecture, and production-minded AI workflows.

Resume Medium GitHub LinkedIn

What I Do

I help turn ambiguous technical and business problems into scoped, testable systems: enterprise AI strategy, customer discovery, proof-of-concept builds, graph-native modeling, LLM workflows, remote team leadership, and deployment planning.

Enterprise AI Architecture

Technical strategy, customer discovery, executive demos, proofs of concept, product feedback loops, and deployment paths for complex data environments.

Hybrid AI Systems

LLM workflows, knowledge graphs, relational reasoning, graph analytics, forecasting, optimization, evaluation records, and model iteration.

Research to Practice

Technical writing, Snowflake ecosystem demos, reference implementations, and research ideas turned into examples people can inspect and adapt.

TastyBytes customer community graph from the RelationalAI and Snowflake graph analytics article

Figure from Bertolani and Siegel 2019 showing ligand docking with and without catalytic geometry constraints

Illustration of a robot reasoning about Bayesian updates and better investment decisions

Experience

2025-Present

Principal AI Solutions Architect / Sales Engineering Lead, RelationalAI

Lead customer strategy, executive-facing demonstrations, and proofs of concept showing how graph analytics, rule-based reasoning, optimization, and AI workflows fit into enterprise data stacks.

2021-2025

Tech Lead / Staff Data Scientist, RelationalAI

Led applied AI projects and a six-person data science team, supported a larger Ph.D.-level organization, and partnered with Sales, Marketing, Product, and partner teams on launches and customer engagements.

2018-2021

Data Scientist, RelationalAI

Built graph and data-science systems for recommendation, next-basket prediction, sales forecasting, taxonomy mapping, semi-supervised clustering, and knowledge graph pipelines.

2012-2018

Ph.D., Physical Chemistry, UC Davis

Developed computational methods for enzyme modeling by combining bioinformatics, structural informatics, and physics-based protein modeling in the Justin Siegel Lab.

Writing and Demos

Latest Medium Articles

From a Bayesian LLM Paper to Blind Predictions in a Week A build note on turning a Bayesian + LLM research idea into a cloud workflow for blind prediction, commitment, and evaluation. Medium What Is AI? Definitions, inspirations, and the people whose work shapes how I think about AI systems. Medium

Selected Demos and Talks

Community Detection using RelationalAI A hands-on Snowflake guide showing graph-native community detection with RelationalAI. Snowflake Quickstart Harnessing Social Graphs A practical walkthrough of using social-graph structure to reveal customer communities and relationship patterns. Snowflake Builders Blog Dovetail Join An explanation of something called a worst case optimal join and why it's very fast on certain problems. RelationalAI Blog Enterprise AI with Snowflake Marketplace A customer-facing session on bringing AI products and data apps into enterprise data workflows. Event Speaker Technical Writing Medium profile for current essays, prototypes, and research-to-practice notes. Medium Profile

Influences

I think AI is about automating things people do. I wrote more about the definitions and researchers that shape my thinking in What Is AI?, and keep a separate list of inspirations.

Research

Before my industry AI work, I built computational tools for enzyme modeling and rational protein design. That research background still shapes how I think about uncertainty, constraints, evidence, and scientific communication.

Published Research

Ph.D. thesis on computational enzyme modeling, geometric constraints, and rational protein design.

Read thesis

Patent

Named inventor on US Patent 10,829,756, "Discovery of enzymes from the alpha-keto acid decarboxylase family."

View patent

Open Source

Former RosettaCommons developer with historical PR contributions; public GitHub includes scientific computing and developer tooling projects.

Rosetta Commons

Stack

Tools and systems I have worked with recently or repeatedly.

Python FastAPI Pydantic Snowflake DuckDB BigQuery PostgreSQL Databricks GCP AWS Neo4j OpenAI API Anthropic Transformers PyTorch React TypeScript Datadog Observe Rosetta

Bio

Short version for intros, applications, speaker notes, and profiles.

Steve Bertolani is an AI systems architect and data scientist working at the intersection of LLMs, knowledge graphs, predictive modeling, optimization, and enterprise data platforms. His background spans computational chemistry, scientific computing, staff-level data science, customer-facing architecture, and production-minded prototyping. He focuses on turning ambiguous technical and business problems into scoped, testable systems that are explainable and useful in real enterprise environments.