Point at a file
S3, GCS, SFTP, a vendor feed or a local folder. CSV, Parquet, gz — schema is detected for you.
Market data infrastructure for quants
Point at a file.
Get a research‑ready dataset.
Map the columns, choose who reads and writes — MD Store lands, normalizes, documents and secures your market data. No pipelines to build, no infra to babysit.
Point at your data
s3://desk-raw/binance/trades_2024-03.parquet
✓ Parquet · 38.1M rows · 5 columns detected
| ts_ms | px | qty | sym | side |
|---|---|---|---|---|
| 1709251200123 | 61234.5 | 0.012 | BTCUSDT | b |
| 1709251200131 | 61234.4 | 0.300 | BTCUSDT | s |
Map columns to the MD Store schema
- ts_mstimestamp ms → UTC ns
- pxprice
- qtysize
- syminstrument → BTC-USDT
- sideside b/s → buy/sell
Dataset
trades.binance.spot
Readers
@quant-research@pm-desk+ add
Writers
@data-eng+ add
Landing…
- Landed raw → raw/binance/trades
- Normalized 38.1M rows → clean.trades
- Deduped 1,204 rows · flagged 3 gaps
- Data card & lineage generated
- Granted read ×2 · write ×1
Dataset ready — md.load("trades.binance.spot")
step 1/4
Concept preview. Screens and numbers on this page are illustrative: MD Store is pre-launch.
How it works
Three inputs from you. Everything else is automatic.
Map the columns
Match source columns to one canonical schema. Units, timezones and symbols convert on the fly.
Choose roles
Pick readers and writers. Permissions are applied everywhere the data lands.
Lands it right
Raw kept immutable, clean layer validated, docs and lineage generated, access granted.
Why MD Store
Everything a data team would build.
Already built.
Build nothing
No pipelines, no infra, no cleaning scripts.
- storage & partitioning
- schemas & versioning
- schedules & backfills
- access control
Zero time on infrastructure
Storage, schemas, schedules and permissions are provisioned for you.
03/01/24 09:00 ESTXBTUSD61,234.5
2024-03-01T14:00ZBTC-USD61234.50
Normalization on autopilot
UTC timestamps, one symbol convention, aligned units and sides.
⌕ trades
trades.binance.spotcrypto · tick2019 → today@data-eng
trades.cme.esfutures · tick2012 → today@data-eng
trades.okx.perpcrypto · tick2021 → today@quant-research
A data catalog, built in
Every dataset searchable by market, frequency, coverage and owner. Find it before you rebuild it.
Know everything
Every dataset explains itself.
trades.binance.spot
Tick-level spot trades for microstructure research.
@data-eng · 2019 → today · 1.2B rows
Docs that write themselves
A data card per dataset: schema, coverage, owner and the idea behind it.
Lineage for every number
Trace any feature back to its files and transforms. Reproduce any backtest.
Research faster
Clean data, one line away.
featuresresearch-ready
cleanvalidated
rawimmutable
A real feature store
Raw data and research features live apart, not in one pile on a shared disk.
import mdstore as md
df = md.load("trades.binance.spot",
start="2024-03-01")A Python library for access
Load any dataset into pandas or polars in one line. No paths, no credentials juggling.
trades.binance.spot38.1M rows
| timestamp | price | size | side |
|---|---|---|---|
| 14:00:00.123 | 61234.50 | 0.012 | buy |
| 14:00:00.131 | 61234.40 | 0.300 | sell |
A data viewer on the web
Browse schema, sample rows and stats right on the site, before writing any code.
Visualizations out of the box
Coverage, gaps and distributions, ready in the viewer or via md.plot() in Jupyter.
Team
Who is building this
Software & Data Engineer
Extensive experience building trading bots, now focusing on Rust development.
Software & Data Engineer
Extensive experience building trading bots, now focusing on infrastructure.
Questions, or want to tell us what your data pipeline looks like? Write to [email protected].
FAQ
Straight answers
Can I use MD Store today?
Not yet. MD Store is pre-launch: the interface on this page is a concept preview and the numbers are illustrative. Join the waitlist and tell us what you work with. We are building around real workflows.
Where would my data live?
Not decided yet, and it is the question we most want your input on. Hosted by us, in your own cloud account or VPC, or fully on-prem? If your data cannot leave your infrastructure, tell us at [email protected]. It directly shapes what we build first.
How much will it cost?
Pricing is not set. We would rather first understand what you spend today, in engineer time and infrastructure, on building and maintaining this internally.
How is this different from DuckDB, Parquet and a few scripts?
Those are excellent building blocks, and for a solo researcher they may be all you need. MD Store is aimed at the layer around them: mapping every source to one schema, validation, a catalog, lineage and access control, so a team does not rebuild it for every dataset.
Stop cleaning data. Start researching.
Join the waitlist for early access.