The LakeHouse
The governed data lake at the core of everything we build. Land it, trust it, shape it, and hand it to your AI agents — from one browser, on any screen.
There is no AI without an IA — an Information Architecture. Your data foundation is the single most important investment you'll make.
Your whole data estate, run from a browser.
The LakeHouse puts storage, quality, engineering, and AI preparation in one workspace. No clusters to keep warm, no tool sprawl to stitch together — and it works on a laptop, a tablet in the meeting room, or a phone on the move.
Data refined in layers — then given meaning
Every dataset climbs a ladder of trust, from the original bytes to business-ready products. Then the Platinum tier goes one step further and turns those products into something AI agents can actually understand.
Bronze
Land & Preserve
The original bytes, never rewritten. Malformed rows are kept aside, not silently dropped — so the audit trail starts on day one.
Silver
Typed & Quality-Gated
Conformed, deduplicated, and passed through quality gates. Upserts and full change history keep records current without losing the past.
Gold
Business-Ready Products
Curated marts, star schemas, and friendly-named semantic tables — published as data products with owners and contracts.
Platinum
Meaning for AI Agents
Ontology, knowledge graph, vector search, and governed agent endpoints — the layer your AI talks to.
Seven zones. One governed lake.
Everything you own lands in one place, in open formats, organized by how much you can trust it.
The lake browser works like a file explorer across every zone — drag in a file or a whole folder, preview it in place, and watch it move from raw to refined. Nothing is overwritten without a trace: every table keeps its version history, and any version can be restored.
- LandingBronze
- RawBronze
- BaseSilver
- EnrichedSilver
- CuratedGold
- SemanticGold
- PlatinumMeaning
- Drag-and-drop upload of files or entire folders, resumable for large loads
- Preview in place — CSV, Parquet, spreadsheets, documents, images, and video
- Time travel — full table history with one-click restore
- Upserts & change history so Silver stays current without losing the past
- Tenant isolation built into every table, not bolted on
- Keyless, identity-based access — no shared storage secrets
The meaning layer your AI agents talk to.
Bronze, Silver, and Gold hold data. Platinum holds what it means. Built only from published data products — each with a schema, an owner, and a quality score — it gives agents a map of your business instead of a pile of files.
Ontology & Glossary
Classes, relationships, and business terms with a draft-and-review flow — plus a summary card agents read first, so they start with context.
Knowledge Graph
Explore entities in 2D or 3D, filter by legend, take a guided tour, and export images or video. The graph reconciles itself as data changes.
Vector Search & Warehouses
Plain-language search over any table, and domain star schemas where every relationship is checked to resolve — meaning and numbers, side by side.
Governed Agent Endpoints
Hosted MCP servers for the graph, warehouse, lake, quality, pipelines, and search — scope-checked, rate-limited, audited, and lockable per tenant. Connect Claude Code, Claude Desktop, or your AI Agent Fleet in minutes.
A score you can defend.
27 rule types across six dimensions, rolled into one severity-weighted score per dataset.
The scoring is deliberately honest. One failing critical rule caps the score, so a serious problem can't be averaged away. A dimension with no rules shows as “not measured” instead of a flattering 100.
- Auto-profiling proposes rules and shows the evidence behind each one
- Rules from a sentence — describe the check, get the rule
- Eleven remediation actions, always applied to a copy, never the source
- Issues explorer drills down to the exact failing row and value
- Quality gates stop bad data before it is published
- 30-day to 1-year trends on every dataset scorecard
From exploration to production in two clicks.
Python notebooks that sit right next to your data — and graduate into scheduled jobs without a rewrite.
Write Python and Markdown in a clean card layout with a full code editor. Charts, tables, images, and interactive widgets render inline. When the analysis is ready, schedule it as a job — the notebook you explored with is the one that runs.
- Choose your compute — serverless, named environments, or a linked Databricks cluster
- Rich outputs — HTML, images, JSON, inline plots, and widgets
- Autosave, variable inspector, and a package panel
- Git folders — clone, commit, pull, and push from the workspace
- Import & export as notebooks, scripts, or archives
- Org-wide libraries curated by administrators
Shape data on a canvas. Ship it as a pipeline.
Eighteen no-code operators — from joins and pivots to AI functions and inline guardrails.
Drag operators onto the canvas and preview up to 1,000 rows at any step. The preview runs the same logic as the scheduled job, so what you see is exactly what production will do. Run the whole flow, or just “run to here.”
- Every join you need — inner, left, right, full, semi, anti, and cross
- AI Functions — classify, extract, summarize, translate, or a custom prompt
- Guardrails apply quality rules inline, mid-flow
- Prepare step — trim, cast, rename, fill, split, parse dates, and more
- AI prompt bar proposes changes; nothing applies until you accept
- Outputs to a table, file, or data product — overwrite, append, or merge
Clean text. Honest signals.
As AI-written content floods your systems, AI Scrubbing keeps it clean, traceable, and fairly judged.
Generated text often carries invisible baggage — zero-width characters, hidden watermark marks, and direction-control codes that break search, joins, and downstream models. AI Scrubbing strips them, normalizes odd spacing, and reports exactly what it changed, column by column.
- Invisible-character hygiene with per-column counts and optional raw-value retention
- Content provenance — reads embedded content credentials where files carry them
- Writing-style signals, clearly labelled as advisory
- Personal-data detection for emails, phone numbers, national IDs, bank and card numbers
- Try it — paste text and see the result before you run a dataset
- Full audit trail of every scrub and every rewrite
Responsible by default. Per-document “AI detector” verdicts are off, because their false-positive rates make them unsafe to act on. Style signals are refused on HR data, and any rewrite needs an admin, an org-level switch, and an audit record.
Realistic data. Zero exposure.
Describe the data you need in a sentence — or mimic a real table — and get related, realistic datasets you can safely share.
A split-screen studio pairs an AI assistant with a live preview. Every suggestion arrives as a clear change card you accept or reject, and every edit is saved to a version timeline you can roll back. Start from a sentence, one of seven industry templates, a real table, or a blank canvas.
- Up to 8 related tables and a million rows each
- Realistic messiness — deliberately missing or invalid values for honest testing
- Four quality dials — fidelity, diversity, validity, and privacy
- Privacy you can measure — exact-copy and nearest-real-row checks
- Documents too — emails, call transcripts, contracts, decks, spreadsheets, and logs
- Tagged & traced — output is labelled synthetic, with lineage
Great models start with great datasets.
Turn your governed data into clean, versioned, training-ready datasets for fine-tuning large language models.
Most fine-tuning projects succeed or fail on the data. The LakeHouse curates it: pick a format, start from a proven example, switch on the clean-up you need, and review what was set aside — with a plain-language “why” for every exclusion. Export train, validation, and test splits ready for standard open-source trainers on the compute you choose.
- Eight formats — chat Q&A, prompt-completion, preference pairs, thumbs up/down, tool use, reasoning, and more
- Proven starting points shaped like well-known public datasets
- Clean-up switches — near-duplicates, low quality, personal data, and toxicity
- Leakage protection keeps test questions out of training
- Versioned exports in common formats, ready to train
- Dataset cards aligned with EU AI Act data-documentation expectations
Compliance built in, not bolted on
Governance runs through every layer. The catalog crawls and profiles your data, proposes column descriptions, and flags personal data; lineage shows every hop from source to job to table; and every change is recorded.
- Data catalog with owners, domains, tags, and classifications
- End-to-end lineage, version history, and restore
- Role-based access — viewer, analyst, engineer, and admin
- Enterprise single sign-on and per-tenant isolation
The data lake at the core
The LakeHouse is the shared foundation beneath every ClearData AI product. Agents read from it, the CRM is fed by it, and every result flows back into it — one governed source of truth instead of three silos.
AI Agent Fleet
Orchestrated agent fleets that reason over Platinum-layer meaning and write their results back to the lake.
LakeHouse
Lands, scores, shapes, and gives meaning to the data — the governed context every agent needs.
ClearData AI CRM
An AI-native CRM fed by agent research and recorded meetings, all landing in the same lake.
Loop Engineering
Self-driving agent loops share one source of state in the lake, so fleets stay coordinated as they scale.
The LakeHouse, explained
Why serious AI starts with information architecture.
A data lakehouse combines the flexibility of a data lake — store anything, structured or not — with the governance and performance of a data warehouse. One architecture serves analytics, operations, and AI from the same governed foundation.
Because there is no AI without an IA — an information architecture. AI agents are only as reliable as the data that grounds them. A lakehouse gives them trusted, current, well-governed data to reason from, which is the difference between cited answers and confident guesses.
A progressive refinement pattern: Bronze holds raw data exactly as it arrived, Silver holds cleaned and conformed data, and Gold holds business-ready data products. The ClearData AI LakeHouse adds a fourth tier, Platinum, which turns published data products into meaning that AI agents can use. Each layer adds trust, so people and agents always know what quality of data they're standing on.
Platinum is the LakeHouse's meaning layer — the tier AI agents talk to. It is built only from published data products that have a schema, an owner, and a quality score, and it adds an ontology, a knowledge graph, vector search, and domain warehouses on top. Agents reach it through governed MCP endpoints, so they reason over trusted data rather than raw files.
Yes, through governed MCP endpoints. Agents such as the AI Agent Fleet, Claude Code, or Claude Desktop connect with a personal key or company sign-in, and every call is scope-checked, rate-limited, and audited. Teams can also turn saved queries and lookups into purpose-named agent tools, which go live only after an administrator approves them.
No — it builds the dataset, which is where most fine-tuning projects succeed or fail. The LakeHouse curates, cleans, and versions training-ready data in eight formats, from question-and-answer chat to preference pairs and reasoning traces, and exports train, validation, and test splits ready for standard open-source trainers on the compute you choose.
It is designed to be, and the LakeHouse measures it rather than assuming it. Every synthetic dataset carries a privacy score based on exact copies of real rows and how close generated rows sit to real ones, alongside fidelity, diversity, and validity. Output is tagged as synthetic with full lineage, so nobody mistakes it for production data.
No. We meet your data where it is — pilots start with the sources one workflow actually needs, and the governed foundation grows with use. Big-bang migrations aren't a prerequisite for value; they're usually what kills it.
Build your data foundation
The LakeHouse is where every AI initiative begins. Let's design the architecture that powers your business.
Get Started