Vigora — Live pilots at two coffee factories

What's running, right now.

Counters refresh every 15 minutes. They reflect what Vigora is observing across both deployments. No hand-tuning, no curated highlights — current operational reality.

The roaster pilot

Site live since: March 14, 2026
Operational days: 47
Roasting sessions observed: 1,047
Signal readings captured: 812,000
Recommendations generated: 142
Machines connected: 1

The soluble factory pilot

Site live since: March 21, 2026
Operational days: 38
Roasting sessions observed: 200
Signal readings captured: 635,892
Recommendations generated: 31
Machines connected: 1 of 6

Aggregate counts only. No tenant identifiers, no signal values, no batch details, no operator names ever leave the sites.

A 22 kg specialty coffee roaster, observed.

Specialty coffee roaster on the production floor; an operator in white lab coat works the control panel; cooling tray of roasted beans in the foreground — Specialty coffee roaster. 22 kg batch capacity. Roasting Light, Medium, Dark, and two Filter Dark recipes daily. Vigora observing since March 14, 2026.

The site

This pilot is a single-machine roastery producing specialty coffee for retail and wholesale customers. The roaster is the production workhorse — eight to twelve batches a day, five recipes in regular rotation, an experienced operator running every shift.

What we connected

Siemens S7-1200 PLC over the S7 protocol. We read nine signals from data block DB200 — bean temperature, exhaust temperature, gas percentage, airflow, drum RPM, ambient temperature, drum temperature, burner percentage, inlet air temperature. Polled every five seconds. Approximately 17,000 readings per machine per day during active production.

What we're learning

The first weeks of operation have been spent calibrating thresholds against the operator's actual practice — finding the band between "Vigora flagged it but the operator was right to ignore" and "Vigora correctly caught a drift." That calibration is the early work. As the system accumulates episodes, the recommendations sharpen.

A six-machine soluble coffee factory, in deployment.

Industrial roaster on a continuous-process soluble coffee production line; large overhead ductwork dominates the upper portion of the frame; ROASTER equipment sign visible on the central machinery — Continuous-process soluble coffee factory pilot.

The site

This pilot is the larger production facility — soluble coffee at industrial scale, with a continuous processing line spanning six distinct stages. Unlike the single-machine roastery, this is where the multi-machine version of Vigora is being proven out. The pilot started with one machine and is expanding stage by stage as profiles are prepared.

What we connected

Today: an industrial continuous-process roaster via Siemens S7 over data block DB21, reading eleven signals (bean temperature, dryer temperature, cooler temperature, furnace flame, furnace temperature, air temperatures across the roaster). The remaining five machines (extractor, evaporator, centrifuge, spray dryer, boiler) have profiles in preparation — connection profiles ready, signal offsets pending operator confirmation. Each machine adds to the picture of what's happening across the line.

What we're learning

A continuous-process line generates different signals than a batch operation. Roaster batch boundaries don't apply when the next machine in the line is always running. We're working with the production team to identify how to track operational episodes that span multiple machines — a single roast batch becomes a downstream input for the dryer, which becomes input for the cooler. Cross-machine learning is part of why this pilot is more interesting than a single-machine deployment alone.

I see what's happening on the line. Vigora sees what's happening across batches. We're starting to catch things earlier — small drifts that I would have missed because I was looking at the next machine.

— Rajesh, Production Lead, Soluble Coffee Factory Pilot

Two deployments, two different problems.

We chose these two pilots deliberately. The single-machine pilot is a focused environment where we can prove the six-stage loop works end-to-end on one machine, with one operator, one set of recipes. The multi-machine pilot is a stress test for the platform's multi-tenant, multi-machine architecture, where the harder problems live.

Together they cover the operational range of our target customer: from a one-machine specialty roaster to a multi-stage industrial line. The platform is the same. The deployment shape is the same. What differs is the depth of process and the number of signals.

We're not optimizing for impressive metrics. We're optimizing for evidence that the architecture holds across the range of factories we want to serve next.

What we're working on at the pilots.

The remaining five machines at the multi-machine pilot come online as profiles are prepared and signal offsets are confirmed with the production team. The deployment work for each new machine follows the same gated, white-glove pattern.

Per-recipe anomaly baselines are coming. Today the system learns thresholds at the signal level. We're moving toward baselines that respect recipe context, so the same signal can have different "normal" depending on what's being produced.

Cross-machine episode tracking. A single roast batch becomes input for the dryer, then the cooler. We're working out how to track operational episodes that span multiple machines without losing the per-machine clarity operators rely on.

Two factories. Two roasters. Six machine types. Live.