QOMBRA AUTO-RUN // PREDICTIVE DECISION-MAKING API
The Predictive Decision-Making API that actually runs the math
Qombra Auto-Run uses tabular foundation models, time series forecasting and causal inference methods under the hood to deliver the same quality as a human data scientist.
One API call. Answers are based on a tabular foundation model that is proven to predict better.
Qombra Auto-Run is for decisions that matter.
The new wave of decision APIs answers in one forward pass. Most of them are known to be poor predictors compared to dedicated predictive models that data scientists use.
The gut call
System One models read the text of your question and return a typed guess in milliseconds: fast, cheap and decent for trivial questions.
But they have never seen your data. Predicting things like customer churn, buying propensities, retention and other aspects of your business require a predictive model.
Running the math
When the answer lives in your data, a decision-making API has to do the work: clean it, preprocess it, fit statistical models to it, and explain the result. Then it answers.
Auto-Run does exactly that on every call: the same pipeline a human data scientist would run, compressed into one request.
Both return typed, calibrated decisions. Only Qombra Auto-Run has actually modeled your data when it does.
How well does it perform?
TabArena. Each score is the share of the gap closed between a naive baseline (0%) and a human data scientist using a SOTA tabular foundation model (100%). Higher is better.
A semantics-only guess is barely reliable. Running the math closes 98.6% of the human data scientist.
What Auto-Run does before it answers
Every call runs the full pipeline. Nothing is skipped, and every step is visible in the response.
Data cleaning
Types are inferred, outliers flagged, missing values imputed. Messy history in, modeling-grade data out.
Preprocessing
Encoding and intelligent feature engineering, with splits built so tomorrow never leaks into today.
Statistical modeling
Tabular foundation models, time series forecasting and causal inference methods, run by the predictive engine: fit to your history, not prompted about it.
Explainability
Every answer arrives with what drove it and how much to trust it: attributions and calibrated confidence from held-out validation.
The answer
A typed, calibrated decision: a class, a number or a probability, grounded in the math that was actually run.
How it works
Three steps between a question and a decision you can act on.
Bring historical structured data
A CSV, a DataFrame or a warehouse table of past outcomes. If this decision has been made before, Auto-Run can learn from it.
Ask the decision
Name the target you want predicted. Auto-Run frames the task: classification, forecasting or a causal what-if.
Get the answer
Calibrated predictions plus the why: attributions and a held-out test score, in the same response.
import qombra qombra.login() run = qombra.auto_run( history_df, ask="Who churns next month?",) print(run.answer)run.predictions.to_csv( "predictions.csv")
run.answer '142 customers are at risk of churning soon' run.accuracy 0.92run.predictions 142 rows x 2 cols run.explain() mean |SHAP| tickets_90d ████████ 0.184 onboarding ████▎ 0.096 nps ███▏ 0.071