Tulan Loom: the agentic delivery platform for Microsoft Fabric.
Six specialized agents that discover, design, build, test, govern, and optimize your Fabric estate, working inside your tenant under the direction of certified Tulan engineers.
Specialists, not a generalist chatbot.
Each agent is tuned for one job in the Fabric delivery lifecycle, with its own tools, guardrails, and evaluation suite. Select one to see what it does.
Discovery & lineage
Crawls your estate and maps every source, pipeline, report, and dependency.
Weeks of discovery interviews become a searchable estate map in 48 hours.
What it does
- Inventories SSIS, ADF, Synapse, Databricks, Informatica, and Power BI
- Builds column-level lineage graphs
- Scores objects for complexity and migration risk
Discovery & lineage
Crawls your estate and maps every source, pipeline, report, and dependency.
Weeks of discovery interviews become a searchable estate map in 48 hours.
What it does
- Inventories SSIS, ADF, Synapse, Databricks, Informatica, and Power BI
- Builds column-level lineage graphs
- Scores objects for complexity and migration risk
Design & standards
Proposes medallion layouts, domain boundaries, and naming conventions.
Design reviews start from a strong first draft, not a blank page.
What it does
- Generates target architecture options with trade-offs
- Applies your standards library to every design
- Drafts architecture decision records for review
Build & convert
Writes pipelines, notebooks, and semantic models to your standards.
Engineers review diffs instead of hand-writing boilerplate.
What it does
- Converts legacy ETL to Fabric Data Factory and Spark
- Generates Direct Lake semantic models
- Opens pull requests with tests and documentation
Testing & parity
Proves data correctness continuously, not just at cutover.
Cutover sign-off moves from gut feel to evidence.
What it does
- Row, sum, and hash reconciliation against source
- Generates data quality tests from profiling
- Blocks deployments that fail parity
Governance & policy
Checks every change against security, privacy, and compliance policy.
Governance keeps pace with delivery instead of blocking it.
What it does
- Detects unlabeled sensitive columns
- Validates workspace roles and RLS
- Keeps Purview catalog and glossary current
Cost & capacity
Attributes every capacity unit and token to a workload and owner.
Finance sees cost per data product, not a single opaque invoice.
What it does
- Real-time CU attribution and forecasting
- Flags runaway Spark jobs and queries
- Tracks AI token spend per use case
Humans direct. Agents execute. Evidence decides.
Loom never merges its own work. Every change is proposed as a pull request with tests, lineage, and a policy report.
Objective
An engineer sets a goal and constraints, like a wave of pipelines and a parity threshold.
Plan and build
Agents plan, generate artifacts, and open pull requests with tests and documentation.
Prove
Sentinel reconciles data; Steward checks policy; Ledger projects cost.
Approve
Engineers review diffs and evidence, then approve deployment through your pipelines.
Migrate with evidence.
Scout scores every legacy object, Forge converts it, and Sentinel proves parity on every run until cutover and beyond.
- SSIS, ADF, Synapse, Informatica, Databricks, and Snowflake sources
- Row, aggregate, and hash-level reconciliation
- Wave planning by value and risk
Build to your standards, every time.
Architect applies your standards library; Forge writes pipelines, notebooks, and Direct Lake models that pass review the first time.
- Metadata-driven ingestion frameworks
- Medallion transformations with tests
- Copilot-ready semantic models
Governance at delivery speed.
Steward checks every change for sensitive data, access, and naming policy, and keeps Purview current automatically.
- Automatic sensitivity labeling suggestions
- Workspace role and RLS validation
- Catalog, glossary, and lineage sync
Know what every workload costs.
Ledger attributes Fabric capacity units and AI tokens to data products and owners, then recommends the fixes worth making.
- Real-time CU attribution and forecasting
- Runaway job detection
- Token spend per AI use case
Enterprise controls, by default.
Loom was designed for regulated industries first.
Runs in your tenant
Loom deploys into your Azure subscription. Source data never leaves your environment.
Least-privilege identities
Each agent uses a scoped Entra ID managed identity with auditable permissions.
Human approval gates
No agent deploys to production. Every change goes through your pull request and pipeline process.
Full audit trail
Every agent action is logged with inputs, outputs, and the approving engineer.
Model choice
Runs on Azure OpenAI and Azure AI Foundry models within your data boundary.
Evaluated continuously
Each agent has a regression suite; releases ship only when evaluation scores hold.
Loom pods vs. traditional delivery.
| Traditional consulting | Tulan with Loom | |
|---|---|---|
| Estate discovery | 4 to 8 weeks of interviews | 48 hours, automated lineage |
| Pipeline conversion | Manual rewrite | 70 to 85% automated |
| Parity testing | Sampled, at the end | Every table, every run |
| Governance checks | Periodic review | On every pull request |
| Cost visibility | Monthly invoice | Per data product, real time |
| Typical team for 1,000 pipelines | 25 to 40 people | 6 to 8 engineers plus agents |
See Loom deliver a lakehouse in an hour.
Join Loom Live on November 4, or book a private demo on your own estate.