v3.4 Iceberg-native semantic layer is live

The data stack collapses
into a conversation.

Lumirai is the AI-native Data Platform-as-a-Service. Ingestion, warehousing, modeling, BI, and ML behind one agent. Your team stops being a tool integrator and starts being a strategist.

Apache Iceberg native
~/lumirai · revenue-warehouse agent · 04:17 · 12 steps
you Pull a clean churn signal from Stripe and Segment, joined to our customers table. Yesterday's data.
agent Found two existing pipelines and a usable join key (customer_id). Inferring schema from stripe.subscriptions.deleted events and Segment identify calls. Materializing as fct_churn_signal in Iceberg.
you Show me what's driving last week's spike.
agent Cross-referencing onboarding cohort, plan tier, and support ticket sentiment
execution trace live
parse_intent42ms
resolve_entities → 3 tables128ms
infer_schema(stripe.events)311ms
propose_join_key87ms
compile_iceberg_plan
materialize
rows scanned 14,221,094
data quality 98.4%
cost $0.041
The problem

Your data stack has 14 tools.
Your data team is exhausted.

Every new question becomes a six-week integration project. Every dashboard becomes another thing to maintain. The modern data stack worked — until it collapsed under its own weight.

01 / TOOL SPRAWL
FVTRNDBTAIRFLSNFLBICTLGQLTY SEMNTREVRSMLOPSORCHMETAGOVNLIN

14 vendors. One question.

Ingestion, warehouse, transformation, orchestration, BI, governance, ML — each in a separate tool, each with its own contract, billing, and on-call.

02 / INTEGRATION TAX

The glue is the work.

Schema drift, broken DAGs, governance gaps between vendors. Engineers spend more time keeping tools talking to each other than answering questions.

03 / SLOW INSIGHT

Months from question to answer.

By the time a request hits production, the question has changed. The board moves on. The team feels the lag, and so do you.

The Lumirai approach

One conversation.
Five capabilities. Zero glue.

We didn't bolt agents onto a warehouse. We built every layer of the platform — ingestion, governance, modeling, BI, and training — to be driven by a single conversational interface.

01 · Ingestion Prompt → Pipeline Connectors written by agents, schemas inferred, lineage automatic.
02 · Management Open governance Iceberg, Delta, Parquet. No lock-in. Cataloging by default.
03 · Visualization Self-tuning dashboards KPIs that adapt to the questions you actually ask.
04 · Modeling Conversational semantics Agents author the model. Humans steer.
05 · LLM Training Sharper with use Models fine-tuned to your data, on your data.
▸ One agent Lumirai core explain last week's churn spike
01 · IngestionPrompt → PipelineConnectors by agents, schemas inferred.
02 · ManagementOpen governanceIceberg, Delta, Parquet. No lock-in.
03 · VisualizationSelf-tuning dashboardsKPIs adapt to your questions.
04 · ModelingConversational semanticsAgents author. Humans steer.
05 · LLM TrainingSharper with useFine-tuned to your data.
▸ One agentLumirai coreAll five capabilities behind one prompt.
How it works

Connect. Converse. Compound.

Three steps. The third is the one nobody else does — your platform gets sharper the longer your team uses it.

01  ·  CONNECT

Connect your sources

describe → infer → materialize

Point Lumirai at your warehouse, lakehouse, SaaS apps, and product events. Agents wire it up; your team reviews and approves.

connect stripe, segment, postgres
ready
02  ·  CONVERSE

Ask in plain language

prompt → plan → answer

Anyone on the team — analysts, PMs, executives — asks the question. Lumirai compiles the model, runs the query, and shows the work.

why is enterprise churn up?
3 cohorts · 1 cause
03  ·  COMPOUND

Get sharper with use

accept → fine-tune → ship

Every accepted query, every correction, every approved metric becomes training signal. Your agent learns your business — privately.

accuracy 87% → 94.2%
Open foundations

Built on what you'd choose if you started over.

Lumirai is opinionated about agents and absolutely unopinionated about storage. Your data lives in open table formats on object storage you control — no proprietary blobs, no exit fee.

your storage, your account, your keys
leave with everything intact, in standard formats
bring any compute engine — Spark, Trino, DuckDB
Apache Iceberg primary table format
Delta Lake interop format
Parquet columnar storage
Apache Spark batch + streaming
OpenLineage lineage standard
dbt Semantic Layer metric portability
What changes

The numbers that move when you collapse the stack.

10× faster pipeline setup avg over 240 connectors · Q4 cohort
80% fewer dashboards to maintain orgs with 50+ legacy dashboards
3.4 days median question → answer down from 6 weeks pre-Lumirai
42% lower data infra spend net of vendor consolidation
Ready when you are

Ready to collapse your stack?

A 30-minute call with our team. We'll map your current stack, show you the consolidated equivalent, and quote it against the bill you're already paying.

No sales pressure · NDA-friendly · 30 minutes