Engineering Solutions That Ship

Installers hand-fill a PDF panel spec, scan it and email it in; sales engineers retype it into a spreadsheet and the ERP, so a quote takes days while installers are trying to win same-day tenders — and the design knowledge that sizes loads, fusing and batteries safely lives only in the engineers' heads.

A self-hosted web configurator built on React Flow (the actively maintained open-source canvas library, chosen because no catalogue tool offers a drag-and-drop schematic editor) walks installers through the same guided sections as the form, lets them drag fused output cards, batteries and extension boxes onto a 2D canvas, and validates load, fusing and battery-runtime maths live. An n8n pipeline prices the BOM only for authenticated approved accounts, pushes the design as a draft quotation into ERPNext for a sales engineer's approval, and generates wiring drawings plus an Ollama-drafted commissioning pack stored in Nextcloud, where the installer reviews, marks up and e-approves the drawings online before paying against the ERPNext quote. Installers can free-issue my company some components and sub-assemblies for us to fit for them. This would require them to mark space on the panel for their gear, including dimensions and locations of input/output terminals, etc...

Hundreds of dual-string VRLA banks across NSW age at block-level rates that nobody can see: discharge tests live on paper forms, impedance surveys pile up in spreadsheets, and the 80%-capacity moment arrives as a surprise — colliding with a supplier who is sometimes eight to ten weeks from having stock. Every year the capex bid for replacements is an educated guess, and the two battery teams route themselves across the state by calendar rather than by which strings actually need attention.

A per-block digital twin of the entire fleet: field readings are captured on a tablet or phone, normalised and temperature-corrected by an n8n pipeline, and written to TimescaleDB (chosen over a vector store because impedance and discharge curves are numeric time-series needing continuous aggregates, and it is a mature open-source Postgres extension) while Neo4j holds the structure — which block sits in which string, at which site, of which generation, with what criticality. An Ollama-hosted model fits each block's impedance and capacity trend to forecast its 80% date; forecasts feed a wave planner that groups whole-string replacements, flags gen-one blocks worth redeploying to lower-criticality sites, checks contracted supplier stock, and drafts purchase orders in ERPNext far enough ahead of the ten-week China lead time. The same forecasts drive a rolling three-year capex view and route the two test teams to the strings closest to end of life first, with the existing DNP3 charger 'battery fail' alarms ingested as an early-warning overlay.

A stock alert in the ERP still turns into a purchase order by hand: someone looks up the preferred supplier, phones for a price, and drafts the order days later, by which time the stockout has already cost a job.

A workflow that watches ERPNext stock levels, checks the preferred suppliers on file first for stock and price, scrapes alternatives with Firecrawl only when the preferred ones cannot fill the order, and drafts the purchase order for a manager to approve. Nothing is sent to a supplier without that approval.

A technician at a remote site needs the manual, the safety procedure and the last diagnostic reading, and usually needs them with both hands occupied. Most of the questions have been answered by a previous technician; the slow path through the full retrieval and sensor stack is taken anyway.

An agent on the edge device that checks a local Qdrant cache of recent diagnostics and common procedures first, answers from it when the match is confident, and only falls back to the live ChromaDB retrieval and sensor fusion on a miss. Voice in, voice out, and it keeps working without a network.

A requirements document is written once and then retyped into tickets by hand, and the two drift apart from the first sprint.

A pipeline that parses the requirements document, generates a structured task tree, and syncs it both ways with OpenProject over MCP, so a status change in either place shows up in the other.

The documents that would make an AI assistant useful are the ones that cannot leave the building.

A self-hosted retrieval system over Nextcloud documents, with Qdrant for the vectors, a local model for the answers, access control checked on every query, and an audit log of who asked what.

Contact-form submissions go unanswered, calculator users leave without a follow-up, and the same person is scored as a new lead every time they fill in a different form, so the CRM holds three half-histories of one prospect.

Every capture point checks ERPNext for an existing contact by email, company and phone before scoring anything. A match is merged and routed to the account executive who already owns it; only a new contact starts the three-email welcome sequence, the scoring and the lifecycle stage changes.

Context evaporates faster than anyone can type it. A name from a site visit, a promise made in a corridor, a battery that will be past its end-of-life date by winter, each is lost between the moment it happens and the next time someone sits at a desk. The tools sold for this assume a structured capture session that rarely occurs. The 'autonomous outreach' products fill a prospect's inbox with mail nobody read before it left. I wanted neither, so I built the opposite.

A sales-intelligence system that takes its orders from one phone, runs on a VPS and a home lab, and sends nothing to a prospect until a human taps a button. At seven each morning a single Telegram message reports what the lead scorer and the equipment monitor did overnight, while a cron outside the workflow engine confirms the engine itself is still alive. Then the day starts, and a thirty-second voice note is transcribed on the home GPU, its names, sites, equipment and standards pulled out by a local model, and the result written three times, to the graph, the vector store and the CRM, each copy pointing at the other two. The follow-up draft arrives on the phone with two buttons and waits up to a day. Silence means skip, and a second, independent lock must read true before any mail leaves, whatever was tapped. Two crawlers hunt for prospects, one over the public web with Anthropic's model and one sandboxed behind a metered proxy, and neither can write to the CRM; their finds wait in a quarantine queue until an Approve in the browser commits them to all three stores at once or rolls the write back. A sequencer wakes every ten minutes, finds leads that changed stage, and picks one of eighteen templates along four axes. On Monday morning the phone offers three episode topics ranked from what prospects have said. One tap later the script is written from sources it can cite, narrated in a cloned voice, animated scene by scene and staged for a Wednesday contact sheet.

A capacity discharge test or a cell impedance survey is run against IEEE 450, 1188 or 1106 depending on the chemistry, with manufacturer thresholds and temperature corrections on top, while the engineer manages arc-flash and hydrogen hazards from a paper checklist. What she noticed on site reaches the maintenance system days later, when the paperwork is filed, if it reaches it at all.

A guided workflow that picks the IEEE procedure for the battery type, trends impedance against the baseline as the readings come in, watches each cell during discharge for reversal or a sudden fall, corrects capacity for temperature, and produces the report to the standard before the engineer leaves site. A Telegram bot takes voice or text notes from any engineer on the crew and feeds them through the same pipeline, so no laptop is needed.

Every team that has tried to let coding agents run unattended has met the same two failures. The agents stop, and nobody notices for hours; or the agents report success, and nobody checks. A pipeline that cannot survive its own restart is a demo, and one that cannot be caught lying is a liability.

A DBOS-backed, Postgres-durable orchestrator runs an eight-agent Claude Code pipeline: planner, builder, reviewer, an injected security stage, tester, deployer and docs, with a spec-author off the main line turning plans into dependency graphs of specs. Every step is checkpointed, every dispatched model call is idempotent against a crash, and every human gate is a durable wait that survives a redeploy. The deployer's claim of success is checked against the remote's file tree before the run may continue, a halt leaves a marker with one of thirteen named reasons, and a reviewer that rejects twice hands the run to a person rather than trying a third time. Four projects run on isolated worker containers from one Compose stack.