Industflow — analytics starter

A small Python package that imports an Industflow data slice into your local MongoDB and runs the headline queries you'd want to start with. The CLI command is mip (short for manufacturing-intelligence package).

About Industflow

Industflow is a manufacturing-intelligence platform built by Industware s.r.o. It connects the production floor — lines, stations, operators, tools — and captures what's actually happening, piece by piece:

• Production tracking. Each product gets a serial number and is followed through every stage of its build (forming → assembly → testing → packing). Multiple stations can contribute to one product across one or more groups.
• Step-level execution. Steps are configured per product type by quality and process operators. Each step instance records timing, the station that ran it, OK/NOK status, captured measurements (electrical resistance, leak/tightness, thickness, vision checks), and a full retry history.
• Shift management. Three shifts per day per line, with automatic expected-vs-actual output reconciliation and downtime classification (planned vs unplanned, machine pause vs operator break, coded reasons).
• Quality & traceability. Defects are coded against a controlled codebook; products can be flagged as suspect for follow-up; first-piece and tester samples get extra checkpoints. A defect-history audit log preserves changes over time.

This repository is the receiver-side companion to a redacted data export shared by an Industflow customer.

Built with uv. No virtualenv to manage, no pip install, no global Python pollution.

Prerequisites

• MongoDB running locally (or anywhere reachable). Default URI is mongodb://localhost:27017/mip. Override with --uri or $MONGO_URI.
• uv — install once: curl -LsSf https://astral.sh/uv/install.sh | sh
• The data drop — a bundle from the sender containing data/*.jsonl, schema.json, SCHEMA.md, RECEIVER.md, and MANIFEST.json. Place the .jsonl files into data/ (see data/README.md).

Quickstart

# 1. fetch dependencies into a local virtualenv (uv handles it)
uv sync

# 2. import the data and create indexes
uv run mip import

# 3. one-page snapshot of what landed
uv run mip summary

That's it. From here, run any of the example scripts.

Example queries

Every script in examples/ is independently runnable and prints a table.

uv run examples/01_outcomes.py            # scrap vs good vs in-flight
uv run examples/02_daily_scrap_rate.py    # day-by-day NOK %
uv run examples/03_loss_by_defect.py      # top defect codes
uv run examples/04_shift_comparison.py    # recent shifts
uv run examples/05_first_pass_yield.py    # FPY per step
uv run examples/06_step_cycle_time.py     # mean / p50 step duration
uv run examples/07_downtime_by_reason.py  # downtime minutes by reason

The example scripts are also self-contained — each one carries its own PEP 723 dependency block, so this works too:

./examples/01_outcomes.py

Customizing

industflow_starter/queries.py holds every aggregation. Each function takes a db handle and returns a list of dicts. Easy to copy into a Jupyter notebook, feed into pandas, or repurpose for a Grafana / Metabase dashboard.

from industflow_starter.db import get_db
from industflow_starter.queries import daily_scrap_rate

rows = daily_scrap_rate(get_db())
# do whatever: pandas, plotly, csv...

Re-importing

mip import is idempotent — it drops each target collection before reloading. Run it again whenever the sender ships a fresh data drop.

What's in the data

Four docs, depending on what you're after — the first two are in this repo, the second two ship with the data drop:

• ABOUT_DATA.md — cookbook-style: model, collections, 9 "try this" snippets, joining patterns, gotchas. Start here.
• AI_IDEAS.md — eight concrete projects you can build with this data + an LLM (shift narrator, anomaly flagging, defect clustering, NL Q&A over operations, trend summarizer…), each with what feeds it, an implementation sketch, and honest caveats.
• SCHEMA.md (in the data drop) — every collection's fields with type, presence frequency, and a description where one was extractable. Open this when you want to know what a particular field means.
• RECEIVER.md (in the data drop) — full domain orientation, joining patterns, indexes, caveats list. The sender's reference document.

Both SCHEMA.md and schema.json are rendered from the same source — schema.json is the machine-readable version (use it from code), SCHEMA.md is the same content laid out for humans.

Troubleshooting

• MONGO_URI must include a database name — your URI is missing /dbname at the end. Default uses mip after the host.
• bson.errors.InvalidBSON: ... — the JSONL files were probably modified in a way that broke Extended JSON. Re-fetch from the sender.
• $median not recognized — needs MongoDB 7.0+. Drop the p50Sec line in industflow_starter/queries.py::step_cycle_time if you're on older Mongo.