SELECTED WORK

The shape of our engagements. Names stay private.

Engagements are anonymized by industry, size and stack, and details are generalised so no client can be identified. Happy to walk through the specifics in conversation.

Anonymized
Industry, company size and stack only. No names, no geography.
Numbers
Rounded and generalised. The stacks and the shape of the work are real.
References
Available in conversation, under NDA.
Healthtech SaaS Series B · ~150 employees Ongoing retainer

An ops agent that answers questions from live warehouse data — without ever seeing a patient record

The ops team wanted to ask questions in plain English and get real numbers back. The compliance team wanted a guarantee that no patient data would reach a model. We built semantic views in Snowflake and an MCP server that satisfies both.

12 semantic views shipped~4 hrs/week back per analyst0 PHI columns exposed to the modelOngoing retainer
Stack SnowflakeSemantic viewsMCPCortexdbt
Read full case →
B2B Fintech ~200 employees 90-day engagement

From brittle SQL pipelines to a Snowflake foundation that ships features in days

A Series-C fintech had outgrown a tangle of Airflow jobs and hand-written SQL. We rebuilt ingestion and transformation on Snowflake + dbt, with tests, docs, and a cost model the team could actually see.

40+ pipelines migrated−35% warehouse cost6 Airflow jobs → 2 dbt models90-day engagement
Stack SnowflakedbtFivetranSnowpipePython
Read full case →
Enterprise SaaS ~400 employees 6-week fixed scope

Cut Snowflake credits by 40% without deleting a single dashboard

The Snowflake bill had doubled in a year and nobody could say why. Six weeks of measuring, consolidating and re-scheduling took 40% off the monthly credits — and left the team with a cost model they still use.

−40% monthly credits31 scheduled tasks → 9 dynamic tablesPayback in week 36-week engagement
Stack SnowflakeDynamic tablesdbtTasks & streamsStreamlit
Read full case →
Retail analytics ~250 employees 10-week engagement

Turning 40,000 free-text customer comments a month into a signal the merchandising team acts on

Reviews, support tickets and survey comments were piling up unread. We used Snowflake Cortex to classify them inside the warehouse, parsed the output into scored, testable columns, and gave the merchandising team a Streamlit app they open every Monday.

~40k comments/month classified3 Streamlit apps in production< $200/month in Cortex spend10-week engagement
Stack SnowflakeCortexStreamlitdbtDynamic tables
Read full case →
B2C Marketplace ~120 employees 8-week engagement

From overnight batch to minutes: an ingestion layer the team stopped babysitting

Every morning started with someone checking whether last night's load had worked. We rebuilt ingestion on Snowpipe, Fivetran and external tables, put tests on every source, and got data latency from six hours down to twelve minutes.

6 h → 12 min data latency180+ dbt tests on sources0 silent failures in the first 60 days8-week engagement
Stack SnowflakeSnowpipeFivetranExternal tablesdbtPython
Read full case →

// MORE CASES PUBLISHED AS ENGAGEMENTS CLEAR NDA REVIEW

// READY TO TALK?

Got a similar problem? Let's see if we can help.

Email or WhatsApp us what you're dealing with. No pitch, no commitment — we'll tell you straight if we're a fit, and if we're not, we'll point you to someone who is.