RELATIONAL TRANSFORMERS · A NEW MODEL ARCHITECTURE

The database
is the context.

Cells become tokens. Relationships carry attention. ASTERISM runs one
pretrained model over the whole graph — then predicts any target, zero-shot.

CONNECTORS143+
ENTITIES RESOLVED8.7B+
RELATIONSHIPS46.2B+
PREDICTIONS / SEC1.2M+
  1. DEPTH 01DATA
  2. DEPTH 02RELATIONS
  3. DEPTH 03CONTEXT
  4. DEPTH 04PREDICTION
THE POPULATION

THE PROMISE

Name the target.
Score the population.

One sentence names the missing cell. One pretrained relational
transformer fills it for every row. The task changes. The model
does not.

SEE A FULL PREDICTION RUN
CHURNWhich customers are likely to churn in the next 90 days? AUROC 0.91 · re-scores as links land
FRAUDFlag fraudulent transactions as they arrive. AUROC 0.97 · scored as they arrive
REVENUEWhat is each account worth over the next 12 months? NMAE 0.24 · per-account estimates
DEMANDHow many units will each SKU sell in the next 30 days? NMAE 0.19 · every SKU, every store
REPEAT PURCHASEWhich customers will buy again this month?
  1. 01cust_88410.92
  2. 02cust_12760.87
  3. 03cust_04130.81
AUROC 0.84 · ranked by probability
MAINTENANCEWhich devices will fail, and when? NMAE 0.27 · days-to-failure, per device

The task changes. The model does not.

A yes or no. A number. A label. If the answer belongs to a row with history, declare the target and score the population.

Which leads will convert this quarter?SALES
What label should this record get?AUTO-LABELING
Which invoices will be paid late?FINANCE
Which claims deserve a second look?INSURANCE
Which patients will miss their appointment?HEALTHCARE
Which shipments will arrive late?LOGISTICS
What will this listing sell for?PRICING
Which tickets should be escalated?SUPPORT
Who is at risk of leaving the team?PEOPLE
Which trial users will upgrade?GROWTH
Which loans will default?CREDIT
Will this study meet its enrollment goal?CLINICAL TRIALS
Which items will be returned?E-COMMERCE
Which suppliers will miss their SLA?PROCUREMENT
Which students are at risk this term?EDUCATION
How many stars will this review give?RATINGS
How hot will the cluster run tomorrow?INFRA
Who is about to hit their usage limit?BILLING

THE PAYOFF · A PREDICTION YOU CAN INTERROGATE

The graph is the evidence. Remove a source, a relation, or a neighborhood. Re-score. Watch the answer move.

MAKE A NEAR MISS

Carve one source, link, or subgraph out of the context and score again. If the metric falls, you found evidence the model was using.

SWEEP THE RELATIONS

Let ASTERISM remove each relation systematically. Rank every source and edge type by lift; tune breadth, depth, and recency on validation.

SEE THE POPULATION

Slice the scores by lifecycle, cohort, source, or time. The model gives you a prediction. The population tells you what it means.

THE METHOD

ONE IDEA TO CARRY AWAY

Attention can follow a join.

A transformer reads a sequence. A relational transformer reads a neighborhood.
Ninety seconds from now, that distinction will feel inevitable.

SLOGAN

“Cells are tokens. Neighborhoods are context.
Attention is the join.”

SYMBOL
RELATIONAL WORLDBOUNDED CONTEXT · CELL TOKENSPOPULATION SCORED
STORY

We want to know whether customer #48291 will churn. Her orders live in one table, her refund in another, her support ticket in a third. A relational transformer does not flatten that world. It gathers it.
Her history enters the context with nearby customers whose outcomes are known. The churn cell is masked. Attention follows the relations. The model writes #48291 → 0.87.
No feature table was assembled. No model was trained for churn. The database supplied the context.

SALIENT IDEA

Cost follows context, not graph size.Every row receives a bounded neighborhood of a few thousand cells. A graph with 46 billion edges costs the same per row as one with 46 thousand. That is the route to scale.

SURPRISE

It has never seen your database.The transformer was pretrained across hundreds of other databases. Yours arrives as context. It reads the schema cold, follows the relations, and predicts zero-shot.

FOUR MOVES. THAT IS THE METHOD.

aREPRESENT THE WORLD

Rows become nodes. Foreign keys become edges. Every cell becomes a token that knows its value, column, table, and place in the graph. Nothing is flattened first.

bGATHER THE EVIDENCE

For each row, the sampler gathers its history and labeled peers into one bounded context. Every fact is sampled as of that row's timestamp, so the future cannot leak backward.

cFOLLOW THE RELATIONS

Cells attend within rows, across links, back through references, and down columns. The join, the aggregate, and the comparison become learned attention paths.

dMASK ONE CELL. SCORE EVERY ROW.

Which customers will churn in the next 90 days?The pretrained model fills the masked target for the whole population. New links arrive. Context changes. The rows score again.

THE FENCE

WHAT A RELATIONAL TRANSFORMER IS — AND IS NOT

NOT A FLATTENED FEATURE TABLE

No giant join turns the database into one frozen matrix. The model keeps the rows, columns, and relationships visible all the way through attention.

NOT A MODEL FOR EACH TASK

Churn does not get one architecture and fraud another. The target changes; the pretrained relational transformer and its learned method stay the same.

A TRANSFORMER OVER STRUCTURE

Values are tokens. Rows and columns provide position. Relations determine which evidence can meet. The database becomes the model's context.

NOW MAKE THE METHOD LIVE

Asterism turns relational transformers
into a database that keeps learning.

  1. 01CONNECT
  2. 02DISCOVER
  3. 03COMPOUND
  4. 04PREDICT
DEPTH 01
01

Bring the world
as it is

Point ASTERISM at the sources
you already run.

Rows become nodes.
Foreign keys become edges.
The schema becomes structure.

No feature table first.
No graph to hand-build.

BROWSE CONNECTORS

SQL / DATABASES

WAREHOUSES / LAKES

EVENT STREAMS / LOGS

APIS / WEBHOOKS

DOCS / EMAIL

CHAT / TRANSCRIPTS

OBJECT STORAGE

SAAS / CRM

150+ CONNECTORS

ASTERISM
CONNECTOR LAYER
CONWAY'S LIFE PLANE X:128 Y:64
DEPTH 02
02

Find the links
you never wrote

The declared schema is only
the beginning.

ASTERISM discovers entities,
foreign keys, joins, and links
hidden across sources.

Every recovered relation
becomes a new attention path.

EXPLORE DISCOVERIES
NEW SCHEMA
orders_container_id ↔
customers_id

confidence: 0.92
type: foreign_key
NEW LINK
product_mentions ↔
support_tickets

confidence: 0.88
first_seen: 2024-04-21
NEW ENTITY
Ticket
confidence: 0.96

source: chat transcripts
NEW JOIN
conversations.user_id↔
billing.customer_id

confidence: 0.94
ENTITIES FOUND8.7B+
SCHEMAS DISCOVERED21.3K
LINKS DISCOVERED46.2B+
JOIN PATHS132M
DEPTH 03
03

Teach it once.
Rewrite the past.

Add a parser today.
Run it over every document,
message, and event already stored.

The extracted facts return
as first-class relations.
Old history gains new links.
Every future context gets richer.

MANAGE PARSERS

PARSERS ADDED OVER TIME

HISTORY
202420232022202120202019OLDER
PARSER V1PARSER V2PARSER V3PARSER V4
+12.4M+18.7M+16.2M+7.3M+9.6M+3.9M+2.2M
AUTOMATIC BACKFILL PROGRESS73%

EXTRACTED FROM CONVERSATIONS

USER
The Snowflake booth was packed.
We had 200+ demos.
ASSISTANT
Did the Snowflake booth raise sales?
SYSTEM (EXTRACTED)
snowflake_booth → sales_influenced
confidence: 0.91
evidence: conversation_27491
date: 2024-04-21
DEPTH 04
04

Ask once.
Score everyone.

Name the target in one sentence.
The pretrained transformer
masks that cell for every row
and fills the population.

Contexts come from cache.
Compute follows rows scored,
not the size of the graph.

RUN A PREDICTION

PREDICTION RUN

Which customers are likely to churn in the next 90 days?


● Model: rt-j · pretrained · zero-shot

  Fine-tuning: none


● Context: BFS over links · 8,192 cells/row

  walk-ranked neighbors · latest first


● Context tuning: grid on validation

  best (breadth, depth) · ensemble ×4


● Output: 2.4M customers scored

  AUC 0.91   calibration 0.98


● Serving

  – 1.2M predictions/sec · p95 38ms

  – embeddings served from cache

  – re-scores as new links land

CHURN RISK
TOP 10 CUSTOMERS LIKELY TO CHURN

0 0.5 1.0
RISK SCORE
RELATIONAL TRANSFORMERSAttention follows the joins
ZERO-SHOTNo training run on your data
OPEN SOURCEApache-2.0. Self-host anywhere.
PRETRAINED AT SCALEHundreds of databases
LIVE PREDICTIONSRe-scored as links land