Platform architecture
Lakehouse and warehouse design on Databricks and Snowflake — Unity Catalog and governance, compute topology, migration paths, and the tuning work that keeps a platform fast as it grows.
Independent data platform consultancy
The data platform work that needs someone who has measured it.
Lakeside Analytics helps engineering and analytics teams design, benchmark, and build on Databricks and Snowflake — then ship the applications and agents that run on top. Evidence first: architecture decisions backed by numbers we produced ourselves, not vendor slides.
Work published by
- Capital One Software
- Plotly
- Databricks SME Engineering
- Virago Analytics
What we do
Four practices, one throughline.
Every engagement starts with measurement and ends with something running in production. No strategy decks that stop at the recommendation.
Benchmarking & cost optimization
Independent, reproducible benchmarks of compute planes and real workloads. We have run the full TPC-DS suite across Databricks compute planes and evaluated Snowflake at 10TB scale — and published the results.
Analytics applications
Full-stack data products: real-time dashboards, billion-point time-series engines, and interactive apps over the warehouse. Rust and Arrow where the performance ceiling actually matters.
AI agents on the data platform
Agents grounded in your own data: knowledge-graph context, governed tool surfaces over text-to-SQL, MCP integrations, and cost supervision so GenAI spend stays legible.
How we work
Measure, then build.
Establish the baseline
Before any recommendation, we measure what your platform does today — query profiles, compute spend, where the time actually goes. Findings are reproducible, so you can re-run them after we leave.
Decide with evidence
Architecture choices get argued on measured trade-offs: cost per query, latency at your concurrency, governance surface. If a benchmark contradicts the conventional answer, the benchmark wins.
Ship and hand over
We build the thing — pipeline, application, or agent — and hand it over with documentation your team can maintain. Engagements are scoped to end.
Selected work
Published, peer-read, and public.
A representative sample. The full index — including the Plotly on Databricks series — is on the writing page.
-
Jobs Classic vs. Jobs Serverless vs. DBSQL: Who wins on TPC-DS?
The full TPC-DS suite — 4,750+ queries — run across three Databricks compute planes via Apache JMeter. Serverless SQL warehouses came out roughly 7× faster than any Jobs compute at equal or lower cost.
-
Scaling agent context with knowledge graphs on Snowflake
How knowledge graphs extend the working context available to AI agents beyond fixed dashboards and flat retrieval, implemented on Snowflake.
-
Visualizing a billion points: Databricks SQL, Plotly Dash, and the Plotly Resampler
Interactive Dash applications over large IoT time-series at scale, using the Databricks SQL connector with resampler downsampling on a Polars backend.
Contact
Tell us what your platform is doing that it shouldn't be.
Benchmarks, architecture reviews, application builds, and agent work. Email is the fastest way in — a short description of the problem is enough to start.