Services
Everything between your sources and your gold layer.
The core service lands and refines your data, and four add-ons put the text inside it to work. Our engineers build, run and monitor all of it in your own warehouse, so there's nothing for your team to operate. Past silver, we can do more when you want it.
01 Core
Ingestion
We connect your databases, SaaS apps, files and event streams, and keep them in sync as often as you need.
- Databases, SaaS applications, files in object storage, and event streams
- Change data capture where the source supports it, so updates and deletes arrive without full reloads
- A historical backfill first, then incremental syncs as often as you need
- Schema changes detected and handled by us, checking with you first when one changes what a table means
- Every sync monitored and retried, and fixed by our engineers when it fails
Lands as One bronze table per source object, with source and load metadata on every row.
02 Core
Bronze & silver tables
Raw history in bronze. Clean, current, queryable state in silver. The same conventions for every source.
- Bronze: every record as received, append-only, so nothing is ever lost or overwritten
- Silver: types cast, duplicates removed, column names standardized
- Updates and deletes merged, so each silver table matches the source as it is now
- Checks on keys, nulls and freshness, so we catch problems before they reach a dashboard
- Consistent naming and documentation for every table and column
Lands as A bronze and a silver table for each source object, in the schemas you choose.
03 Add-on
PII redaction
We find personal data in structured columns and in free text, and redact or mask it before it spreads.
- Detection of names, emails, phone numbers, addresses, account numbers and other identifiers, including inside free text
- A policy per column, agreed with you: redact, mask, hash, or replace with a consistent token
- Applied where you choose: before data lands, so raw values never reach bronze, or on the way into silver, with bronze restricted
- A findings table recording what was redacted, where, and when
Lands as Redacted silver columns, plus a findings table for review and audit.
04 Add-on
Document ingestion
PDFs, Office files, web pages and scans, parsed into text and structure your warehouse can query.
- Documents from object storage, shared drives and SaaS apps, synced alongside your other sources
- Text extraction that keeps the structure: headings, tables, lists and page numbers
- OCR for scanned pages and images
- Metadata for every file, and a content hash so unchanged files are skipped
Lands as A documents table and a sections table, each linked back to the original file.
05 Add-on
RAG pipelines
Chunks and embeddings that stay in step with the source, ready for retrieval-augmented generation.
- Chunking suited to the content: by section for documents, by message or field for tickets and notes
- Embeddings from the model you choose, regenerated only when the source text changes
- Source, date and access metadata on every chunk, so retrieval can filter on them
- Stored in your warehouse, ready for its vector search
Lands as Chunk and embedding tables, each row traceable to its source record or document.
06 Add-on
AI enrichment
Turn free-form text into typed columns: sentiment, topics, entities, intent and summaries.
- Ready-made enrichments for sentiment, topic, entities, language, intent and summaries
- Enrichments specific to your business, such as churn risk or product mentioned: you describe what you need, and we build and test it
- Runs on new and changed rows only, so the same text is never processed twice
- Each value recorded with the model and prompt version that produced it
Lands as Typed enrichment columns next to the text they came from, ready to group, filter and join.
01 Every pipeline
What every pipeline includes
- Setup, monitoring and fixes by our engineers, with nothing for your team to watch
- Permissions on your data sources managed by us, or kept in-house by your team
- Schema changes, source API changes and backfills handled for you
- A note from us when something affects your tables: what happened, and what we did
- Tables in your own warehouse, under your roles and permissions
- Full history in bronze, so silver can be audited or rebuilt from any point
- The same naming and metadata conventions for every source
- Documentation for every table and column we write
02 Beyond silver
More hands, when you want them.
Bronze and silver are our core. These are extras: work past silver that the same team can take on when you want it.
- Gold layer modeling. Marts, metrics and business-ready models built on silver, in dbt or the tool you use, to definitions your team signs off.
- Machine learning models. Forecasting, churn, scoring and other models trained on your own tables, with predictions written back to your warehouse.
- Data science. Analysis and experiments on your data, from one-off questions to the numbers behind a big decision.
- AI knowledge layer. The definitions, documents and context that AI assistants and agents draw on to answer questions about your business correctly, developed and managed by us.
- Data visualization. Dashboards and reports in the BI tool you already use, built on your gold models.
03 For clarity
A few things Ingot is not
- Not software you run. There's no app to learn and no pipeline to watch: our engineers do that.
- Not a one-off project. We don't build pipelines and hand them over; we keep running them.
- Not a warehouse. Ingot writes to the one you already have.
- Not a BI tool. Your dashboards live in the tool you already use, whoever builds them.
- Not the owner of your business logic. Gold is yours to define, even when we build it.
- Not a black box. Every silver row can be traced back to the raw record in bronze.
Security review
Bringing your security team in early?
Good. We'll walk them through the architecture: what connects to what, where data is processed, who at Ingot can access what, how personal data is found and redacted, and what the AI add-ons can and can't see. Ask for it when you book.
04 FAQ
Questions data teams ask
What exactly lands in bronze, and what in silver?
Bronze holds every record as the source sent it, append-only, with metadata on when and how it arrived. Silver holds the current state of each object: typed, deduplicated, with updates and deletes applied, and with personal data redacted if you use that add-on. Both are ordinary tables in your warehouse.
Who manages permissions on our data sources?
Whichever you prefer. We can manage the permissions on your data sources for you, or your team can keep them in-house and give us read-only access to each source. Either way, you’ll know exactly what access Ingot has.
What happens when a pipeline breaks?
We fix it. Our engineers monitor every sync, so a failure is our job, not an alert in your team’s inbox. If it affects your tables, for instance by making data late, we tell you what happened and what we did about it.
What happens when a source changes its schema?
We handle it. New columns are added to bronze and silver as they appear. When a change could break something downstream, such as a type change or a dropped column, we deal with it in silver, and check with you first if it changes what a table means.
How fresh is the data?
As fresh as you need, set per source. Sources that support change data capture can sync continuously; others sync as often as their APIs allow.
How are deletes and backfills handled?
Deletes from sources that report them are recorded in bronze and applied to silver. When a backfill is needed, we load the source’s history into bronze and rebuild silver from it, so reprocessing never means pulling from the source again.
How accurate is the PII redaction?
It varies with the kind of identifier, the language and how messy the text is, so we measure it on a sample of your own data before you rely on it. Every redaction is logged, so you can review what was caught.
Which models do the AI add-ons use?
You choose, per enrichment and per embedding pipeline. Every output records the model and prompt version that produced it, so a change of model is visible in the data rather than hidden in it.
Next step
Hand us the pipelines.
A 30-minute call about your sources, your warehouse and the text you want to put to work, with a quote to follow. Or send a question and a person will reply.