Data Scientist, Patterns

Tellwise, Inc.; San Francisco (CA); On-site; USD 120,000 - 190,000; Full time

Job summary

Tellwise, Inc. is seeking a data scientist to build models that distinguish recurring from one-off transactions and detect creep and anomalies. You will set thresholds for warnings and review results weekly against labelled histories, documenting inaccuracies.

You will prioritize precision of warnings over recall and communicate findings to non-analysts with clarity. This role focuses on reliable, defendable signals in transactional data.

Qualifications

• 4+ years in applied data science, ideally on transactions or time series.
• Ability to write for non-analysts and communicate clearly in a few minutes.
• Willingness to challenge the product and say when it’s wrong.
• Focus on precision of warnings over recall.

Responsibilities

• Build models that classify recurring versus one-off, detect creep and anomalies.
• Set thresholds for warning signals and determine what constitutes 'enough' certainty.
• Review output weekly against labelled histories and document what is wrong.
• Keep false alarms rare to maintain trust in warnings.

Skills

Time series analysis

Anomaly detection

Model development

Threshold tuning

Job description

When Tellwise says a subscription is creeping up, or that a charge was off-pattern, that claim has to come from somewhere defensible. You own where.

What you would own

• Build the models that classify recurring versus one-off, and detect creep and anomalies.
• Set the thresholds: how sure is sure enough to show someone a warning.
• Review output weekly against labelled histories and file what is wrong, specifically.
• Keep false alarms rare. A warning nobody needed erodes trust in the one that mattered.

What we are looking for

• Four or more years in applied data science, ideally on transactions or time series.
• You can write for a reader who is not an analyst and has four minutes.
• You are comfortable being the person who says the product is wrong about this.
• You care about precision on warnings more than recall.

How hiring runs

A thirty-minute call with a founder, then a paid work session on a real problem from our backlog (three to four hours), then two conversations with people you would work alongside. Four steps, usually inside two weeks. We tell you where you stand at every one, and we reply either way.