The dataset you need, with a source behind every record.
Market maps, company universes, product catalogs, research corpora, eval sets. We build one-off datasets from public web sources, document how, and hand over data you can defend in a meeting or a model card.
When we're a fit, and when we're not
A good fit if…
- You need a dataset that does not exist off the shelf.
- The answer must be traceable: investors, clients or a model card will ask where it came from.
- You are building an eval set, RAG corpus or fine-tuning set from niche or multilingual web sources.
- You may want it refreshed later, without rebuilding from scratch.
Use something cheaper if…
- A commercial data provider already sells the exact dataset. Buying it is faster.
- You need expert human labeling at scale (RLHF, preference data). Specialist human-data firms do that.
- A few hundred rows will do. A browser agent and an afternoon are enough.
What you get
Deliverables you can check, not hours you have to trust.
Written spec
Universe definition, fields, inclusion rules and edge cases agreed before collection starts.
The dataset
Typed, de-duplicated and validated, in CSV, Parquet, JSON Lines or straight into your warehouse.
Provenance columns
Source URL, retrieval timestamp, extraction method and an observed vs. derived flag on every record.
Method note
How the universe was built, what was excluded and known gaps, written for a reader who wasn't in the room.
AI-ready packaging
For AI work: chunking, near-duplicate removal, language tags, licensing signals per source and held-out splits.
Refresh option
Turn the sprint into a managed feed later; the extractors are already built.
Records you can trace
Each record says where it came from and whether a value was read from a page or derived from others.
| Entity | Field | Value | Kind | Source |
|---|---|---|---|---|
| acme-robotics | founded | 2019 | observed | acme.example/about |
| acme-robotics | employees_band | 51–200 | derived | 3 sources |
| acme-robotics | hq_city | Austin, TX | observed | registry.example/… |
| nordwind-labs | products | 4 | observed | nordwind.example/products |
Illustrative example. Your schema is agreed with you.
What teams use it for
Market maps
Every company in a niche with size, product and positioning, for strategy, investment or sales planning.
Product catalogs
Structured catalogs across brands or retailers for assortment and pricing analysis.
Eval sets
Domain-specific question and answer sets from authoritative web sources, with held-out splits.
RAG corpora
Clean, chunked and deduplicated knowledge bases from docs, regulations or industry sites, with freshness tracking.
Agents build. A QA gate decides what ships.
- 1
Scope
We map the sources, the fields and the most reliable way to get each one, and send you a fixed quote.
- 2
Agents build
Coding agents write extractors against your schema and test them on live pages. A person reviews the change.
- 3
QA gate
Every run passes deterministic checks, then an independent model spot-checks a sample. Failures go back, not out.
- 4
Run and monitor
Plain code runs the feed on schedule. When a site changes we are alerted, an agent drafts the fix, we review it.
Questions we get
Something else? Ask us directly.
How long does a Data Sprint take?
Most sprints deliver in one to two weeks after the spec is agreed. Very large universes or many heterogeneous sources take longer; the spec states the date.
Do you sell training data for frontier labs?
Not at frontier-lab scale. We build niche, documented corpora and eval sets for teams building domain models, RAG systems and agents, where provenance and freshness matter more than raw volume.
How do you avoid duplicated or contaminated data?
Near-duplicate clustering, URL and content hashing, and held-out splits built before any filtering that could leak. Each record keeps its cluster ID so you can audit decisions.
Can you include personal data?
By default we build company- and content-level datasets, not lists of private individuals. If a project needs personal fields, we scope that explicitly with you first.
Get a free 100‑row sample
Tell us what you need to track. Within 2 business days you'll get a sample, the sources we'd use and a fixed price, or an honest note that an off-the-shelf tool will do.
- Sample and fixed quote in 2 business days
- Source URL and timestamp on every row
- No commitment, no sales sequence
Prefer email? [email protected]