Kodatek is a technology consulting practice that pairs deep technical expertise with something rarer — the ability to communicate it. We turn complex data into clear business intelligence and confident decisions, partnering with teams across financial services, enterprise, and high-growth companies to deliver machine learning, AI, and analytics that move the business forward.
A decade of experience across
What we do
From the first business question to a model running in production — across data science, AI, engineering, and the dashboards that make it usable.
Predictive models that earn their place in production — fraud and risk detection, recommendation systems, forecasting, and classification, with the validation and governance to back them.
RAG pipelines, agentic workflows, and document-automation systems built on GPT-4 and open models — with LLM-as-a-judge evaluation so quality is measured, not assumed.
Reliable pipelines and cloud data platforms that turn messy source data into trustworthy, query-ready foundations for analytics and ML.
Executive-ready reporting that people actually use — clear KPIs, governed metrics, and dashboards that drive daily decisions instead of collecting dust.
Client-facing apps and internal tools that put your models and data in front of the people who need them, built to be fast, maintainable, and on-brand.
Translating business goals into technical roadmaps — A/B testing, measurement, and pragmatic guidance that keeps data work tied to revenue and outcomes.
Selected work
A sample of engagements across financial services, retail, and enterprise IT. Client names are generalized for confidentiality.
Led the end-to-end design of an enterprise anti-money-laundering detection model (XGBoost) — architecture, feature governance, selection, and validation — plus an automated engine that turns model output into multi-page investigator case reports backed by SHAP-driven transaction evidence.
fraud detection
case reporting
Architected a RAG agentic system on GPT-4 that ingests customer, party, and transaction data and produces structured narrative case reports for AML investigators — with pattern-detection prompts that surface anomalies and an LLM-as-a-judge metric scoring completeness, accuracy, and clarity.
manual reporting
quality control
Designed and deployed an enterprise product-recommendation engine across 12 airports and 60+ vendors, with an A/B testing strategy and rollout roadmap that stood up the company's first ML-driven personalization capability.
incremental monthly revenue
60+ vendors
Directed an end-to-end migration of the reporting stack from AWS QuickSight to Google Looker, with automated pipelines, dynamic dashboards, and a KPI governance framework used by C-suite and department heads for daily decision-making.
full migration
daily decisions
Built a PCA + logistic-regression classification system to prioritize IT tickets in real time, transforming a historically manual workflow into an automated, ML-powered pipeline.
resolution accuracy
turnaround
Engineered an automated validation and profiling system for datasets with 300+ columns and 10M+ rows, producing intuitive visual diagnostics for non-technical users and removing a major QA bottleneck.
QA turnaround
300+ columns
How we work
Senior, hands-on delivery with the structure of a larger firm — and none of the overhead.
We start with your business objective, not the technology — mapping the decision the data needs to drive, the constraints, and what success measurably looks like.
A clear technical plan: data sources, modeling approach, architecture, and a validation strategy — sized to your timeline and team.
Iterative delivery of models, pipelines, dashboards, or apps — with checkpoints so you see progress and can steer along the way.
Validation against the metrics we set, documentation, and a clean handoff so your team can own and extend the work with confidence.
Contact
Whether it's a model to ship, a pipeline to fix, or a dashboard your leadership will actually use — tell us what you're trying to do and we'll tell you how we'd approach it. No obligation.