Software and statistics,
built well.

DataThink is a small studio of engineers and statisticians. We build web applications, ship R packages, and apply AI carefully.

tidyversepandaspolarsduckdbStanNext.jsTypeScriptPostgresscikit-learnggplot2rmarkdownDatabricksQuartoSnowflakedplyrFastAPIPrismaTailwind
app.datathink.io / dashboard
Q3 Forecast · weekly
$1.846M +18.4%
1W1M1Q
MAE
0.012
Coverage
94.1%
0.946
Web Engineering

Software, built to last.

Marketing sites, internal tools, customer portals, dashboards, full product UIs. TypeScript, Next.js, Postgres, modern design systems. DataThink builds web software that serves your users today — and that your team can maintain and extend for years to come.

See the web practice
CRANdatathink/forecastr
v2.4.0
DESCRIPTION
Package: forecastr
Title: Hierarchical Forecasting Models
Version: 2.4.0
Authors@R: DataThink
Depends: R (≥ 4.2)
Imports:
tibble, dplyr, rlang,
cmdstanr, posterior
License: MIT
✓ R CMD check — 0 errors, 0 warnings
forecastr.R
forecast_hb <- function(data, horizon = 12) {
# fit hierarchical model
fit <- cmdstanr::sample(
stan_model,
data = stan_data,
chains = 4,
iter = 2000
)
}
R Package Development

R packages that outlast the analyst who wrote them.

DataThink writes, tests, and maintains R packages — tidyverse-compatible, S4 where it matters, documented with roxygen, shipped to CRAN. The kind of statistical tooling that quietly runs inside dozens of teams' pipelines, year after year.

See our packages
AUDITrun · 0938.json
reviewed · 2026-05-14
Prompt
Summarize Q3 cohort retention and flag any segment with CI lower bound below the target.
Model
claude-opus-4.7 · temp 0.2 · seed 42
Sources
4 internal datasets · cited inline
Tool calls
read_csvfit_cohort_modelvalidaterender_report
Validation
held-out check · 94.1% match vs. ground truth
Human review
Approved by E. Glenn
sha256 · 7a4f…e21breproducible · ✓
Responsible AI

We use AI when it actually helps.

LLMs help us read faster, draft tests, scaffold migrations, and pressure-test analyses. We reach for them when they earn their keep — and not when they don't. Either way: auditable prompts, validated outputs, and no model decides anything alone.

See our AI principles
How we work

The DataThink guarantee.

Engineers write the code.

Not handed off, not abstracted behind an account manager. The same people on the kickoff call ship the code.

Reproducible by default.

Every deploy, pipeline, and analysis runs the same way the next time you touch it. No 'works on my machine.'

We document the boring parts.

Onboarding notes, run-books, and decision logs so your team owns what we built when we leave.

Available when it matters.

Working remotely with teams in 14 states. Headquartered in Rexburg, Idaho — on your hours, not ours.

Tell us what you're trying to ship.

DataThink takes on a small number of engagements per quarter. Bring the project — we'll scope it honestly, whether that's a sprint of frontend work or a year of statistical modeling.

Start a project

Bring us the part that's hard.