Beta · live demo behind this page

Your herd already writes the data.
Now it answers back.

The Barnboard Data Platform turns every milking you log into governed metrics, open APIs, and a per-cow milk forecast — running on the same small computer as Barnboard itself, with nothing sent to anyone's cloud.

Try the live demo Read the documentation

90-second silent walkthrough — the story, the PIN gate, and the live app with its forecast panel.

Q What is the avg daily yield per cow?
23.16 L/cow/day
receipt · lineage gold.agg_milk_cow_daily ← silver.milk_logs · owner data-designer · v1
Q Show avg daily yield per cow, by breed
HF cross 26.64 · Murrah 24.95 · Jersey cross 21.47 · Gir 21.20 · Sahiwal 18.72 …
every number carries its SQL and its lineage — answers you can audit

The problem

A dairy's records usually die in the notebook they were born in.

Numbers without meaning

Litres get written down twice a day, but "is the herd doing well?" still gets answered by gut feel. Averages computed by hand disagree with each other, and nobody can say where a figure came from.

Data locked in

Records trapped in one app — or one notebook — can't feed a vet's spreadsheet, a co-op's reports, or next year's loan application. Getting data out (and back in) shouldn't be a rewrite.

Problems found late

A cow that slips 20% over three days is telling you something — mastitis, heat, feed. By the time the drop is obvious in the monthly total, the cheap intervention window is gone.

What you get

🥇 A governed data pipeline

Every table flows through a bronze → silver → gold refinery with lineage recorded at each step. Metrics are defined once, governed always: the platform physically refuses to sum a ratio up a grain — the classic spreadsheet mistake that quietly corrupts averages.

📈 A forecast on the dashboard

A per-cow model (trained on your own records, standard-library Python, no GPU, no cloud) projects each cow's next 7 days. The dashboard shows the herd's expected week; a sustained drop against forecast flags the cow — in the demo it catches the seeded decliner with zero false alarms.

🔓 Open contracts, not lock-in

Everything the farm enters is available as CSV and a versioned JSON API with a schema endpoint any consumer can program against. Export, re-import, back up — the round-trip is tested, and the whole database is one portable file.

🔔 Alerts that earn trust

Withholding (DO-NOT-TANK), yield-drop early warning, and likely-heat detection — a cow on her 21-day cycle showing the tell-tale one-day milk dip gets flagged in time to breed her. Every alert needs two signals to agree, so quiet mornings stay quiet.

📊 A price outlook, honestly built

Next year's ₹/L estimate on the dashboard: 35 years of India's official milk price index give the direction, your own collection records give the level. Backtested before it was allowed on screen.

🌍 Benchmarks that fit your herd

Cited benchmark yields for the breeds Indian farms actually keep — Gir, Sahiwal, Red Sindhi, Murrah buffalo, HF and Jersey crosses — sourced from NDDB, ICAR and peer-reviewed studies, loaded right next to your own numbers for honest comparison.

Architecture

One machine, one stack, four layers. Data flows down; every layer is inspectable and every answer carries a receipt.

Barnboard app milkings · breeding · sales Reference data open datasets · checksummed Breed benchmarks NDDB · ICAR · cited per row Medallion pipeline — bronze → silver → gold typed landing · cleaning · modeled facts · lineage on every step Governed semantic layer metrics as versioned definitions + guard Models 7-day yield forecast · drop flags · revenue Open contracts JSON API · schema endpoint · CSV round-trip Served inference predict API · dashboard forecast panel Runs anywhere Python runs — laptop · farm PC · small VPS · (container: on the roadmap)
Answers flow back up the same path they came down — with lineage attached.

Feature matrix

What was asked for, what was actually built, and what's still ahead. A matrix that's all green is a matrix nobody believes — this one is honest.

RequirementWhat exists todayStatus
Interoperable — data usable by other systems Versioned JSON API (/api/v1/…), schema endpoint, CSV export and import (round-trip tested), one-file database with a rehearsed backup drill Built
Extensible — grow without engine changes A new metric is one JSON definition; governance guard blocks unsafe rollups; extension path documented and demonstrated Built
Governed pipeline with lineage Bronze→silver→gold with per-step lineage, PII exclusion at ingest, re-runs proven bit-identical Built
Free web data, properly researched 16-source inventory with licenses; 3 datasets stored locally with sha256 checksums; gaps recorded openly (no open per-cow desi-breed data exists anywhere) Built
Breed benchmarks incl. desi breeds Six breeds with per-row citations (NDDB, ICAR, peer-reviewed), loaded into the gold layer Built
Per-cow yield forecasting 7-day damped-trend model; beats the naive baseline on a held-out week (MAE 1.25 vs 1.33 L) Built (3 weeks of demo data — margins will grow with history)
Early-warning drop detection Deviation-vs-forecast flags; catches the seeded decliner, zero false positives; half-logged days excluded so a missed PM entry never fakes an alarm Built
Revenue projection Month-end estimate from month-to-date rate; 8.1% holdout error on demo data Built (demo-scale; needs multi-month history to trust for planning)
Inference served in the product Predict API + dashboard forecast panel; model artifacts versioned and hot-reloadable with zero downtime Built
Automated tests 23-test suite over contracts, models, auth and rate-limiting — all passing Built
Runs on bare metal, supervised PM2 + systemd boot unit; kill-and-resurrect drill passes — all six services survive a power cycle Built
Container deployment Docker image builds and serves — login and dashboard verified running inside the container Built
Milk-price outlook from market data FAOSTAT India milk price index (35 years) gives the direction, your own sales give the level — a dashboard tile shows next year's ₹/L estimate, backtested (drift beats naive, 4.6% vs 6.9% error) Built
Heat (estrus) detection 21-day cycle + one-day milk-dip rules flag "likely heat" on the dashboard; zero false positives by construction (both signals must agree) Built
Governed metrics API Every governed metric served with its receipt — definition, owner, version, lineage, and the exact SQL (/api/v1/metrics) Built
Operations that run themselves Nightly database backups (14-day retention, restore rehearsed) · nightly pipeline + model retraining · CI on every code change Built
Encrypted access on the farm network HTTPS on the shared lanes with secure cookies (self-signed until a domain brings a public certificate) Built
Multi-farm / co-op analytics Design drafted, deliberately parked by owner decision Parked

Roadmap

  1. The lactation curve takes over on its own — a curve-based forecaster already competes in every nightly retraining contest (it has beaten the trend model on a multi-month backtest); it ships automatically the night it wins outright on your real data.
  2. Narrated demo video — the hero walkthrough gains a voiceover.
  3. A public certificate — when a domain arrives, the self-signed HTTPS on the farm network upgrades to a browser-trusted one.
  4. Co-op view — parked by decision; the design is written and waiting.

Deployment notes

On-premises (the lane you're looking at)

The whole platform — app, pipeline, models, this site — runs on one modest machine under a process manager. Four pinned Python dependencies for the app; the data and model layer is standard-library only. Your data never leaves the building: models train and infer locally, and the reference datasets are fetched once and checksummed.

Cloud twin (when you want it)

The same codebase runs on a small VPS with no code changes — the database is a single file you copy, and the demo seed makes a walkable synthetic herd on first boot. A deploy/twin compose file exists from the app's earlier package; the data platform's container lane ships when the Docker image is verified (see the matrix — it's marked honestly).

See it with real motion

The demo is the actual product on a synthetic 12-cow herd — log a milking, watch the forecast panel, pull the API. Ask us for the access PIN if you don't have it.

Enter the live demo Documentation