E03: Can AI Turn Engineer’s Notes into a Diffuser Specification? | The Specification Test

Wastewater plants collect enormous amounts of operational data — flow, dissolved oxygen, air flow, energy, ammonium, temperature and more. Much of it is stored but rarely analysed in depth.In this episode, AI receives a synthetic year of hourly SCADA data containing several deliberately hidden problems: sensor faults, outages, unit changes, energy-meter resets, calibration spikes, daylight-saving-time issues and misleading correlations caused by rainfall.The challenge is not simply to make graphs. It is to determine whether AI can distinguish a genuine operational insight from bad data — and produce findings that an operator or engineer could actually investigate.

A calculation can look perfectly professional and still be completely wrong because of one unit, decimal separator or hidden assumption.In this episode, we give AI ten wastewater-engineering traps involving flow conversions, mass loads, nitrogen and phosphorus reporting, population equivalents, gallons, pressure, decimal commas and even daylight-saving time.Some problems require a calculation. Others require something more important: recognising that there is not enough information and stopping to ask a question.Can AI detect the trap — or confidently produce the wrong answer?

The new EU Urban Wastewater Treatment Directive introduces major changes for treatment requirements, timelines, energy, micropollutants and investment planning. But turning a long legal document into something an engineer can actually use is another challenge.In this episode, we test whether AI can convert Directive (EU) 2024/3019 into a practical obligations register: what must be done, who it applies to, the relevant plant-size thresholds, the conditions that trigger each requirement, and the deadlines.Then comes the critical part: checking whether the AI has interpreted the articles, thresholds and dates correctly. Because one wrong deadline in a regulatory summary can eventually become a wrong investment decision.

Aeration calculations contain dozens of assumptions, conversions and equations — making them a perfect test for AI.In this episode, AI works through a synthetic wastewater aeration design involving oxygen demand, AOTR, SOTR and required air flow. Then it gets a second challenge: reviewing a calculation containing deliberately seeded errors.Can AI recognise incorrect assumptions, unit problems and calculation mistakes — or confidently continue with the wrong numbers?A practical test of how AI could be used as a second pair of eyes for wastewater process calculations, while keeping the engineer responsible for the final design.

What happens when you give AI a page of engineering notes and ask it to write a tender specification?In this episode, ten synthetic design notes for a fine-bubble diffuser system are turned into a technical specification covering design conditions, materials, guarantees, testing, installation and documentation.We compare different AI workflows and examine one of the biggest risks of AI-assisted specification writing: requirements that sound professional but were never actually provided by the engineer.The real question is not whether AI can write a specification. It is whether every requirement it writes can be traced back to an engineering decision or a reliable source.

Can AI read a wastewater treatment tender and identify the questions an engineer should ask before submitting a bid?In this episode, we test AI on a public wastewater aeration tender. The task: find missing design data, contradictions, unresolved cross-references and potential deviations, then turn them into clear clarification questions.But generating questions is only half the job. Every quotation, page reference and technical claim has to be checked against the original tender before anything could ever be sent to a client.A practical look at where AI could help wastewater bid engineers and where engineering judgement still matters.