Accepting new engagements
StatGazer audits and builds forecasting, backtesting, and risk models for hedge funds, asset managers, and quant desks. You get reproducible code, validation, methodology, and handoff materials that stand up to IC, risk, and audit review.
01section What we solve
Most engagements start with one of these — a model or a result someone needs to trust before they act on it.
The numbers are strong on paper, but you cannot tell whether the strategy works or whether future information leaked into the test. Sizing into it means betting that a result you cannot fully reproduce is real.
We reproduce the backtest from raw, point-in-time data and trace every input for look-ahead, alignment, and survivorship — so you know which part of the result is signal and which is leakage before capital is at risk.
The model was fit on one environment and quietly stops working when volatility, correlations, or the underlying regime shift. By the time the drawdown shows it, the damage is done.
We model the regimes explicitly — state-space, regime-switching, and structural-break methods — and stress the strategy across environments, so you understand how it behaves when conditions turn rather than assuming they will not.
The model drives decisions, but the assumptions, validation, and limitations live in someone's head or a notebook no one else can run. When a risk committee or an auditor asks, there is nothing to hand them.
We document the model the way review demands — methodology, validation, limitations, and reproducible code — and close the governance gaps, so the result and the reasoning travel together to the people who sign off.
The forecast is a single number, presented as if it were certain, with no sense of how wrong it could be. Decisions get made on a point estimate the model itself cannot support.
We build forecasts that state their uncertainty — calibrated intervals and distributions, validated out-of-sample — so you can size and hedge against the range of outcomes instead of a single guess.
02section Engagements
Each engagement is scoped, fixed in writing before we start, and delivered as reproducible work your team keeps — an audit, a build, a review, or a private cohort.
E.01
An independent review of a strategy or model before you size it up — to find leakage, fragile assumptions, and results that will not reproduce.
Deliverables
OutcomeYou learn whether the result is real before capital depends on it — and get a documented, independent opinion you can put in front of an IC or a risk committee.
E.02
A forecasting or risk model built from your data, validated out-of-sample, and delivered with the uncertainty stated — not hidden.
Deliverables
OutcomeYou get a model whose behavior you understand and can explain — including where it is weak — instead of a black box you have to take on faith.
E.03
A review of the pipelines, data flow, and deployment around your models — so the system is reproducible, monitored, and ready for audit.
Deliverables
OutcomeYou move from a model that works on someone's machine to a documented, reproducible, monitored system you can defend to oversight.
E.04
Practitioner-led training for your team, built around your stack and the problems you actually face — not a generic course.
Deliverables
OutcomeYour team leaves able to spot the mistakes that sink models — and to build and review work to a higher standard on their own.
03section Process
Four steps, with a deliverable at each one. You always know what you are getting and when.
We define the problem, the data you have, the decision it feeds, and what defensible has to mean for your IC, risk committee, or auditor. We agree on scope, constraints, and success criteria before any work begins.
You receiveA written scope and engagement plan — objectives, data and access requirements, milestones, and a fixed or scoped fee.
We examine what exists — the model, the backtest, the assumptions — or run the research the build depends on. We surface failure modes early: leakage, look-ahead, regime sensitivity, and the gap between in-sample fit and a live decision.
You receiveA findings memo — what holds, what does not, the methods we tested, and a ranked list of issues with recommended fixes.
We implement the model, pipeline, or fix in code, with validation built in rather than added after. Every result is reproducible from data lineage to output, and reviewed against the criteria set in scoping.
You receiveReproducible code and notebooks, a validation report, and the model or system itself, documented for review.
We hand the work to your team so it runs without us — walking through the code, the assumptions, the monitoring, and the limits. The goal is that your people can defend, maintain, and extend it.
You receiveHandoff documentation, a walkthrough session, and where applicable a monitoring plan and the artifacts needed for governance or audit review.
04section Proof
Every engagement is built to be inspected: reproducible notebooks, methodology memos, validation reports, and handoff documentation. Where client work is confidential, we show the standard through redacted examples and synthetic-data artifacts.
What every engagement produces
Code and notebooks that regenerate every result from raw data to output — no manual steps, no unexplained numbers.
A written account of the methods, assumptions, and choices behind the work, so the reasoning travels with the result.
Out-of-sample tests, leakage and look-ahead checks, and known limitations — written for a reviewer, not just for us.
The setup, structure, and operating notes your team needs to run, maintain, and extend the work without us.
Where a model goes to production: what to watch, what should trigger review, and how to tell when it has drifted. Provided where applicable.
Representative engagements
Model & backtest audit. A quant or PM team about to allocate to a strategy whose backtest looks strong. Independent review of the strategy's data handling, backtest construction, and assumptions — checking for leakage, look-ahead, survivorship, and overfitting before capital is committed.
Forecasting & risk model build. An investment or fintech team that needs forecasts with honest uncertainty around them. Building a probabilistic forecast, scenario design, or VaR / Expected Shortfall model, with validation and documentation built for review.
ML infrastructure review. A data-driven team whose models work in research but are hard to trust in production. Review of the pipeline that feeds and serves models — lineage, features, evaluation, deployment, and monitoring — to find where results can silently degrade.
See a sample findings memo built on synthetic data — the format and standard of the deliverable. A reproducible sample notebook is available on request, and detailed case studies are in preparation. Engagement details and references are held under NDA; we do not publish client names, data, or results without written permission.
05section Education
We teach the same econometrics, statistics, and machine learning we apply in engagements — math-first where it matters, hands-on throughout. The work runs through the Private Quant Training Cohort; the areas below are the ground it covers.
06section AI & Data Engineering
The infrastructure that carries research into production: data pipelines, feature stores, and the deployment, monitoring, and evaluation that keep models accountable. Engineered for reproducibility and audit — and, where a client genuinely needs it, extended to on-chain data and settlement.
Streaming and batch pipelines, feature stores, and reproducible training and deployment. Versioned data and models, evaluation gates, and monitoring — so what runs in production is the thing that was validated.
Model integration, retrieval, and evaluation pipelines built for production use — measured against task-specific metrics, monitored in operation, and accountable when they drift.
For clients who need it: ingestion and analysis of on-chain data, and security-first smart-contract work with auditable, upgrade-aware patterns. Offered on request, not as a headline.
07section Founder
Evgenii Azarov
Data Scientist · Financial Engineer · Educator
StatGazer was founded by Evgenii Azarov, a data scientist and financial engineer who works at the intersection of econometrics, statistics, and machine learning. He combines research depth — time-series and causal methods, forecast evaluation, model validation — with the engineering to put it into production, including shipped software and iOS applications. Alongside the consulting work, he teaches these methods to practitioners, which keeps the firm's standard concrete: research that can be implemented, and engineering that is quantitatively honest. Every engagement carries his direct involvement rather than being handed off.
Reach the founder directly at hello@statgazer.com.
08section FAQ
Every engagement starts with a scoping call, then a written scope and fee before any work begins. From there we run four steps — scoping, audit or research, build, and handoff — and you receive a deliverable at each one. Most engagements are scoped as a fixed piece of work; longer or embedded research runs on an agreed cadence.
Yes — we sign an NDA before reviewing confidential data, and we are glad to work under yours. Where possible we work inside your environment or on a minimal extract, retain only what the engagement needs, and return or delete data on request at the end. We do not reuse client data to train models for anyone else.
You own the deliverables produced for your engagement — the code, models, notebooks, and reports. We retain our pre-existing methods, tooling, and general know-how. Specific ownership and license terms are set in the engagement agreement so there is no ambiguity later.
Pricing is scoped per engagement. After the scoping call we send a written scope with a fixed fee where the work is well-defined, or a clear rate and estimate where it is open-ended. We do not quote a price before we understand the problem, and we do not bill for work outside an agreed scope without agreeing it first.
We reply to enquiries within one to two business days. Turnaround on the work itself depends on scope and is agreed in writing before we start — we would rather commit to a date we can hold than to one that sounds fast.
For scheduled training, you can cancel or reschedule up to ten business days before the start date for a full refund or credit; after that, fees may be partly retained to cover reserved time. For self-paced access, terms are stated at the point of purchase. Full terms are in the Refund Policy.
09section Contact
Bring us a model, a backtest, or a forecasting problem you need to defend. We will tell you the shortest honest path to a result you can stand behind — and whether we are the right team for it.
We reply within 1–2 business days · Direct line hello@statgazer.com
What happens on the scoping call
Training & courses
A separate track from consulting. If you want a paid program such as the training cohort, request access and we will send terms and a checkout link once it is live.
Request program access