DATA CHARTS PLAN EXP RESULTS
Path A · Insights

Ask anything about your data

Counts, averages, segments, charts — answered in plain English the moment you load the file.

"how many companies?" · "plot ARR" →
Path B · Train

Train a model

Tell OctOpus what to predict and the agent runs experiments, picks the winner, ships you a model.pkl.

"predict who hits $10M ARR" →
Connect your data
21 connectors · or drop a CSV / Parquet / ZIP
CSV / Parquet
file upload
BigQuery
warehouse
Snowflake
warehouse
Databricks
warehouse
Redshift
warehouse
Postgres
database
MySQL
database
MongoDB
database
Amazon S3
object store
Google Sheets
SaaS
Notion
SaaS
ClickHouse
warehouse
yc_recent_batches.csv
100 companies · 17 cols · batches S22 → S24
company_namebatchindustryarr_usdpost_money_valuation_usdmom_growth_pctmonths_since_demo_day
YukonSecS23Security541,15342M14.212
NovaOSW23Dev Tools532,89038M9.818
SpireAPIS22API Infra463,72455M11.424
EchoLabsW24AI / ML421,224120M22.56
IrisPayS22Fintech403,95068M8.124
HaloLabsS23AI / ML253,50472M17.812
ZeroBioW23Biotech235,96128M6.218
FluxSecS24Security204,23815M28.43
ArcEngineS24Dev Tools193,45412M31.03
KaiTechW24AI / ML191,92385M19.26
ARR distribution across industries
100 companies · median $23.6k · skew right
AI / ML
$284k
Fintech
$231k
Dev Tools
$190k
Security
$148k
API Infra
$112k
Biotech
$64k
Research plan program.md
targetreached_10m_arr (binary derived from arr_usd)
taskclassification
metricROC-AUC, holdout-validated
splitgroup k-fold by batch (so later cohorts can't leak into training)
featuresindustry, founders, technical_founders, employees, last_round_usd, runway_months, mom_growth_pct, has_paying_customers, months_since_demo_day
experiments1) CatBoost · 2) LightGBM (Optuna 30 trials) · 3) TabPFN · 4) stacking ensemble
guardrailsleakage detection, holdout outside agent workspace, scrubbed env vars
Research loop
running 4 experiments
1
CatBoost · baseline
AUC 0.82
2
LightGBM · Optuna ×30
AUC 0.85
3
TabPFN · zero-shot
AUC 0.87
4
Stacking ensemble · 3 base learners
AUC 0.89
WINNER
Stacking ensemble · validated winner
0.89AUC · holdout
Octopus
Hi, I'm OctOpus, your AI Data Scientist. Drop a dataset and tell me your goal — I can train a model or answer questions about your data.
21:29
Octopus
Got it. I can see your yc_recent_batches.csv dataset. Profiling it now…
21:30
Octopus
GOALWhat's your goal?
Skip
21:30
Data insight (Ask OctOpus)
21:30
Octopus
Insights · FocusWhat should I inspect?
Skip
21:30
Summary
21:30
Octopus

Here's what I see in this dataset:

It covers Y Combinator companies from recent batches — 17 columns capturing identity, funding, growth, and team composition.

Key financial signals available: arr_usd, last_round_usd, post_money_valuation_usd, runway_months. mom_growth_pct is the velocity signal.

Team data: founders, technical_founders, employees capture team composition.

Stage spread: has_paying_customers is a binary flag separating pre-revenue from revenue-generating companies.

Targets worth modeling: post_money_valuation_usd, runway_months, or a derived reached_10m_arr binary.

21:31
what's the total arr and the top companies?
21:31
Octopus
Across 100 companies, total ARR is $7,056,039. Top by ARR is YukonSec at $541,153 — there's a notable spread between leaders and the rest.
company namearr_usd
YukonSec541,153
NovaOS532,890
SpireAPI463,724
EchoLabs421,224
IrisPay403,950
+ 95 more rows
21:31
show me arr distribution across industries
21:31
Octopus

ARR distribution across industries

AI/ML Fintech DevTools Sec API Bio SaaS Other
arr_usd ranges $0 → $541k, median $23.6k, mean $110.2k (right-skewed). AI/ML leads with $284k median.
21:31
train a model that predicts $10M arr
21:32
Octopus
Plan ready. Group-fold by batch so later cohorts can't leak. Open the Plan tab to inspect, or type go.
21:32
looks good, run it
21:32
Octopus
Running 4 experiments — CatBoost baseline, LightGBM with Optuna, TabPFN zero-shot, and a stacking ensemble of the strongest base learners.
21:33
Octopus
Done. Best model: stacking ensemble at AUC 0.89 (holdout). model.pkl ready in the Results tab.
21:35
OctOpus is idle
Ask anything about your data or model
1 / 9 Pick a path
Pricing Try it on your data →