Get in touch



For decades, data scientists have been forced into an opaque and wasteful question-first data science workflow by black-box, question-oriented AI/ML. We're changing that.


Ask and answer questions instantly and transparently

Map the information in your data

Discover which variables are mutually predictive; know which questions you can answer before you ask them.

_ = c.heatmap('depprob', cmap='gray_r', vmin=0, vmax=1)

Predict anything given anything

Free yourself from defining inputs and output. Once you've fit a model, you can predict anything in your dataset given anything else (or nothing at all). And Redpoll gives you uncertainty information by default. Always know how confident the model is in its prediction.

pred, uncertainty = c.predict(
  'Country_of_Operator', given={'Launch_Mass_kg': 5000}

('USA', 0.09967501286209424)

# Apply confusing conditions
pred, uncertainty = c.predict(
        'Launch_Mass_kg': 5000,
        'Launch_Site': 'Taiyuan Launch Center'

('China (PR)', 0.4270984931675438)

Identify the sources and magnitude of uncertianty

When you predict using today's ML technology you get a number. Take it or leave it. But Redpoll gives you uncertainty information, and allows you to visualize multiple sources of uncertainty so you know when to hold, when to fold, and when to collect more data and what type to collect.

    given={'Class_of_Orbit': 'MEO'},

Identify anomalies and errors

Because Redpoll builds an intuitive model of the process that generates your data, it's easy to intuit when data do not adhere to the model. When you build a model in Redpoll, you get data QA/QC for free.

# 5 most surprising satellites by Orbital Period
c.surprisal('Period_minutes') \
    .sort_values(by=['surprisal'], ascending=False) \
Period_minutes surprisal
Wind (International Solar-Terrestrial Program) 19700.45 13.884460
Spektr-R/RadioAstron 0.22 9.523201
Interstellar Boundary EXplorer (IBEX) 0.22 9.494666
Integral (INTErnational Gamma-Ray Astrophysics Laboratory) 4032.86 9.182491
Geotail (Geomagnetic Tail Laboratory) 2474.83 9.143796
Chandra X-Ray Observatory (CXO) 3808.92 8.922031
XMM Newton (High Throughput X-ray Spectroscopy Mission) 2872.15 8.910115
Tango (part of Cluster quartet, Cluster 2 FM8) 3442.00 8.341243
Rumba (part of Cluster quartet, Cluster 2 FM5) 3431.10 8.310921
Samba (part of Cluster quartet, Cluster 2 FM7) 3430.28 8.308759

Simulate synthetic data

Generate secure synthetic datasets that maintain the same statistical properties as the original data or create de-biased datasets. Increase the speed of innovation by reducing the barrier to collaboration.

# generate hypotheticals
values = c.simulate(
    ['Launch_Mass_kg', 'Country_of_Operator'],
    given={'Class_of_Orbit': 'GEO'},

# synthesize a new dataset
full_data = c.simulate(

Find knowledge gaps

Ever wonder where you need to improve your data — where either messiness or lack of data causes uncertainty in your prediction? With Redpoll, it's easy.

import numpy as np

lifetimes = np.linspace(0, 40, 200)
uncertainties = []
for lifetime in lifetimes:
    _, u = c.predict(
        given={'Expected_Lifetime': lifetime}

Stream-native means you never retrain

Edit existing values

Because Redpoll has already mapped the statistical structure of the data, it knows which parts of the structure are affected by updated data.

    row='Sicral 1B',

Scales with your business: add new rows or columns

Data collection is a continual process, so Redpoll accepts new data continually. And if you start collecting new fields, that's fine too. Redpoll is designed to grow with your data.

    'Landsat 9',
        'Expected_Lifetime': 9,
        'Class_of_Orbit': 'GEO',
        'Launch_Mass_kg': 2711

import pandas as pd

  pd.read_csv('functional.csv', index_col=0).is_functional,

Lifelong learning

Redpoll is a lifelong learning system that thrives where traditional ML/AI systems fail. Redpoll's streaming capabilities allow it to

  • serve as an automated QA/QC platform that can detect anomalies before they enter the dataset
  • detect and adapt to data drift so your models are as dynamic as the world
  • learn new concepts as more data become available

Get in touch

Redpoll is deliberate about the organizations with whom we choose to partner. If you’re interested in working with us, please fill out the brief form below and we’ll set up a time to connect!