We build and improve machine learning & AI decisioning systems.
We help enterprises identify and execute high-impact use cases where better algorithms and models directly improve revenue, profit, and efficiency.
Applied Machine Learning for Effective Decisioning Systems
We help enterprises uncover high-impact use cases where better algorithms and models directly improve revenue, profit, and efficiency.
Custom Decisioning Systems
We design and launch end-to-end machine learning systems tailored to your business logic. From fraud detection to dynamic pricing, we build production-grade solutions that deliver measurable ROI.
Model Rescue & Audits
We diagnose why existing models are failing, whether from data drift or stale algorithms, and execute the engineering fixes needed to restore accuracy and trust.
Platform Modernization
We implement modern infrastructure that makes models easier to update, fully reproducible, and compliant with governance standards.
Cost & Speed Optimization
We refine algorithms to cut processing time and optimize cloud infrastructure, ensuring your system remains profitable as you grow.
We build what works, engineered by people who’ve done it before.
Principal‑level engineers and researchers who tackle your hardest algorithmic challenges and deliver systems engineered for outsized, verifiable ROI.
We Build What Works
We focus on ML and AI systems that directly move revenue, risk, and efficiency, designed around your data, workflows, and governance. The result: robust decisioning systems that are production‑ready, observable, and built to last.
Deep, Proven Expertise
Half our team holds PhDs in math, statistics, or computer science. The rest are senior engineers who’ve led ML platforms and decisioning systems at Microsoft, Capital One, Lyft, and Amazon. We’re comfortable owning the hardest technical and architectural pieces end‑to‑end.
Impact Without Overhead
We embed with your team to deliver senior‑level ML and AI outcomes: fraud losses avoided, conversion lifted, cost per decision reduced, without the cost and long‑term commitment of building a full internal team.
Driving Enterprise ROI Through Applied AI/ML Engagements.
We partner with fintech, financial advisory, and consumer‑tech clients to build and modernize machine learning and AI systems that drive financial and operational performance and deliver high ROI.
Fintech Lender | Payment Fraud Detection Rebuild
Challenge
A leading fintech lender’s fraud detection model had degraded after years without retraining. Model drift, fragile data infrastructure, and slow retrain cycles reduced fraud catch rates and increased false positives.
Solution
Modernized the fraud data platform, sharpened key signals, and deployed a new, easy-to-retrain payment‑fraud model into production.
Impact
- $3.5M in projected annualized savings
- 81% increase in economic value per payment over the legacy model
Fintech Lender | Payment Fraud Detection Rebuild
Challenge
A leading fintech lender’s fraud detection model had degraded after years without retraining. Model drift, fragile data infrastructure, and slow retrain cycles reduced fraud catch rates and increased false positives.
Solution
Modernized the fraud data platform, sharpened key signals, and deployed a new, easy-to-retrain payment‑fraud model into production.
Impact
- $3.5M in projected annualized savings
- 81% increase in economic value per payment over the legacy model
M&A Advisory | LLM‑Driven Market Intelligence Enrichment
Challenge
A collision‑repair M&A advisory needed to expand its market‑intelligence database and keep prospect data current across tens of thousands of U.S. businesses. Previous operations involved a slow and manual analyst workflow.
Solution
Designed and implemented an LLM‑driven data‑enrichment system that automates market discovery and shop‑level data collection, turning a manual research process into a scalable intelligence pipeline.
Impact
- 20% growth in verified prospect list
- 90%+ accuracy versus human evaluation
- 30+ analyst hours per quarter saved
M&A Advisory | LLM‑Driven Market Intelligence Enrichment
Challenge
A collision‑repair M&A advisory needed to expand its market‑intelligence database and keep prospect data current across tens of thousands of U.S. businesses. Previous operations involved a slow and manual analyst workflow.
Solution
Designed and implemented an LLM‑driven data‑enrichment system that automates market discovery and shop‑level data collection, turning a manual research process into a scalable intelligence pipeline.
Impact
- 20% growth in verified prospect list
- 90%+ accuracy versus human evaluation
- 30+ analyst hours per quarter saved
Retail | Recommendation System Modernization
Challenge
A consumer‑engagement platform’s recommender relied on highly-biased popularity-based heuristic and failed to capture personalization across campaigns and markets.
Solution
Rebuilt the recommendation engine with an item‑to‑item collaborative‑filtering approach, modern modeling, and full production rollout, validated through controlled A/B experimentation with a global retail partner.
Impact
- 10–20% lift in purchase conversion across 100M+ messaging events
- Stable, redeployable recommender powering personalized campaigns at scale
- Strategic differentiation: positioned the platform to compete with enterprise‑grade personalization engines
Retail | Recommendation System Modernization
Challenge
A consumer‑engagement platform’s recommender relied on highly-biased popularity-based heuristic and failed to capture personalization across campaigns and markets.
Solution
Rebuilt the recommendation engine with an item‑to‑item collaborative‑filtering approach, modern modeling, and full production rollout, validated through controlled A/B experimentation with a global retail partner.
Impact
- 10–20% lift in purchase conversion across 100M+ messaging events
- Stable, redeployable recommender powering personalized campaigns at scale
- Strategic differentiation: positioned the platform to compete with enterprise‑grade personalization engines
An all-technical team of senior ML scientists and engineers.
Staff+ talent from Lyft, Capital One, Amazon, and Microsoft. Half hold PhDs in math, stats, or CS.
DJ Rich
Howe Wang
Michael Ernst
Tanya Roosta
Alex Hasha
Neal Fultz
Matthew Willian
background
Matthew is a data and infrastructure engineer focused on scaling reliable, cost-efficient machine-learning and analytics platforms. He has led data platform and governance efforts at Plaid and Metronome, supporting critical financial workflows and enterprise launches, and works with growth-stage companies to improve reliability, MLOps, and infrastructure efficiency. His work is grounded in hands-on system design and operational execution. Matthew holds a Bachelor’s degree in Mathematics from Bowdoin College and completed visiting studies at New York University.
background
Neal is a data scientist specializing in forecasting, optimization, and credit risk modeling. He has built and maintained machine learning systems for alternative lending, insurance, and large-scale optimization in both startup and research environments, including ZestFinance, System1, and Google. His work combines statistical rigor with hands-on data engineering and model deployment in production settings. Neal completed doctoral studies in Statistics at UCLA and holds a Master’s degree in Information Systems from UC Berkeley and a Bachelor’s degree in Computer Science from Washburn University.
background
Alex is a machine learning and data science leader with deep experience in credit, fraud, and enterprise ML platforms. He has built and operationalized automated training, validation, and monitoring systems at companies including Mission Lane, Capital One, and Persefoni, with direct ownership of high-impact credit decisioning models deployed at scale. His work sits at the intersection of modeling rigor, regulatory constraints, and production reliability. Alex holds a PhD in Mathematics from New York University and a Bachelor’s degree in Mathematics from MIT.
background
Tanya is an applied AI leader with extensive experience building and deploying large-scale machine learning and generative AI systems. She leads applied science teams at Amazon across Search and conversational AI, and previously built regulatory-grade and production ML systems at Moody’s Analytics, the Federal Reserve, and fintech startups. Her work emphasizes evaluation, safety, and the reliable translation of advanced research into customer-facing systems. Tanya holds a PhD in Electrical Engineering and Computer Science from UC Berkeley, along with graduate training in statistics and financial engineering.
background
Michael is a senior software and data engineer with deep foundations in formal systems and large-scale production software. He has spent several years building and maintaining complex systems at LinkedIn and leading engineering efforts in regulated technical domains, with a focus on correctness, clarity, and operational reliability. His background brings formal rigor to practical system design and implementation. Michael holds a PhD from UC Irvine and a Bachelor’s degree from Harvey Mudd College.
background
Howe works at the intersection of machine learning/ai systems, data engineering, and product execution. He has led data products supporting ML- driven and complex decision systems at Lyft and founded LLM-powered data platforms focused on real-time market intelligence and investment analytics. His work emphasizes translating complex data into automated, reliable operational decisions. Howe holds an MBA in Finance from Wharton and a Master's in Computer Science from Johns Hopkins.
background
DJ Rich works with clients to translate business constraints and objectives into algorithmic solutions and deliver them into production. Before co-founding True Theta, he spent five years at Lyft as a machine learning research scientist, developing pricing, forecasting, and planning algorithms. He also worked in the hedge fund industry modeling risk and credit portfolios. DJ holds a Master’s in Financial Engineering from UC Berkeley.
Ideas from True Theta.
Applied perspectives on machine learning, decision systems, and the real economics of engineering.
Mutual Information
Our YouTube channel on the mechanics of modern machine learning systems, where models, data, and incentives meet. We cover topics like model collapse, evaluation, and scaling laws with a focus on what practitioners need to ship and operate better systems.
Latest from our YouTube Channel
From latest videos
What Happens When All Training Data is AI Generated?
From latest videos
Why Don't AI Agents Work (Yet)?
True Theta Substack
Longer-form essays on model governance, infrastructure, and machine learning strategy, drawn from real enterprise deployments. Concrete frameworks and playbooks you can adapt to your own decision systems.
Latest from our Substack
From latest blog post
Should You Retrain That Model? (Part 3)
From latest blog post
AI Inference Platforms: A Practical Guide
Let's talk about your model.
Whether you're debugging a model or planning a new build, we can help.