Please Check Our Design With ChatGPT. I Mean It.

The Philippines now sits around sixth in the world for ChatGPT use. Sixth, out of every country on earth. So it was only a matter of time before it started happening to us, and this year it happens almost every month: a client takes our proposal, pastes it into a chat window, and comes back with questions.

Some of my competitors hate this. I have heard them complain about it at supplier meetings. My position is the opposite. Please check our design with AI. I mean it. If a free chatbot can talk you out of a design I gave you, then the design was wrong and you just saved both of us two years of arguing beside a tank that does not work.

But I want to tell you what it will actually catch, and what it will miss, because I have now watched it do both.

What it catches: the contradiction inside your own documents

Earlier this year we were asked to look at a sewage treatment plant design for a commercial building here in Cebu. Another company had already issued the design. The client was not asking me to redesign it. They were asking me whether it would work.

I read their documents before I read the design. Buried in the specification was a line saying the wash bay for their delivery trucks would produce water at around 1,485 mg/L of suspended solids. Two pages earlier, the same document described the plant's influent as domestic sewage. Domestic sewage in this country runs about 200 to 250 mg/L TSS.

Both numbers were written by the same team, in the same document, for the same plant. Nobody had noticed. The design had been sized for the second number and the plant was going to receive the first one.

You do not need thirteen years in wastewater to catch that. You need to read the document and ask a simple question: does this number agree with this word? That is exactly the kind of work a language model is good at. It never gets bored on page 40. It does not assume the engineer who wrote page 12 already checked page 14.

So if you are going to use AI on a proposal, do not start by asking whether the technology is good. Start here: "Read all of my own documents and list every place where two numbers contradict each other." In my experience that single prompt finds more real problems than any technical debate about which process is better.

What it misses: everything that was never written down

Now the other half of that same project.

Once we knew the wash water was real, the obvious answer was a closed-loop recycle system for the wash bay. Clean, impressive, and expensive. I killed it. Not because the engineering was wrong, but because we could not get the client to tell us how many trucks they actually wash in a day, and I was not going to price a system around a number nobody could give me.

What we did instead: catch the grit and free oil right at the wash slab, hold one day of wash water in a buffer, and trickle it into the plant at about 0.4 cubic metres per hour over twenty hours so it never arrives as a shock. Then let the biological stage handle what is left. Once we ran the numbers, the plant as designed could absorb up to 10 cubic metres a day of pre-treated wash water without growing by a single tank, because 55 cubic metres a day spread over the day is 2.75 cubic metres per hour, and the hydraulic design already allowed 3.00.

The recycle system that would have cost the client more than double stayed in the proposal, honestly, as an option worth building only above roughly 10 cubic metres a day of washing.

No AI gave me that. It could not have. That answer came from a number that existed only in our own hydraulic calculation, combined with a judgement about a client who was never going to count their own trucks. None of that is on the internet.

The time it was confidently wrong, and I pushed back

This is the part I think matters most, and it is the part people never post about.

On a different design I was working on ammonia removal using advanced oxidation. I was told, flatly and with confidence, that catalytic ozonation cannot remove ammonia. It sounded authoritative. It was also the kind of statement that would have quietly changed my whole design.

I did not accept it. I asked for every paper to be pulled and re-checked, and the picture that came back was not the same as the summary. Catalytic ozonation over manganese catalysts does remove ammonia, at 45 to 100 per cent, most of it converted to nitrogen gas. The catch is that all of it was done in clean water, at pH between 9.5 and 11, with powdered catalyst, and there is no full-scale plant anywhere running it. So for my project it was useless, but for the wrong reason from the one I was given.

And when we kept reading, the mechanism that does work at scale showed up. Ozone followed by biological activated carbon, tested back in 1989 at 0.5, 8 and 500 cubic metres per day, giving 86.8 per cent ammonia removal against 61 per cent for carbon alone. The authors credited the biology growing on the carbon, not the ozone. That paper changed my design. The confident one-line answer would have changed my design too, in the opposite direction, and I would have been the one explaining to a client why the plant was not nitrifying.

The lesson I took: the first answer you get is a summary of the loudest consensus. It is a starting point, not a finding. Ask it for the paper. Ask for the flow rate the study was run at, the pH, the year, whether it was a beaker or a plant. Then go read the paper yourself. Every time I have done that, the truth was more specific and more useful than the summary.

Why I actually want clients doing this

My first serious project, years ago, was a food processing plant making hams, hotdogs and beef jerky. I underestimated the emulsified fats and oils in their process water. I specified a grease interceptor when what that stream needed was dissolved air flotation. On top of that, the domestic and process wastewater had been combined into one line, which pushed us into stricter compliance parameters than either stream would have faced alone. The plant sat on the second floor. When the odour started, the complaints did not stay with the client. They went to the Barangay, then the City Health Office, then all the way to the DOH. The full account is here.

We fixed it. We retrofitted fixed-film media into the aeration tank, we did not bill that client again until the plant passed and worked, and we were on that site at all hours, including through the holidays. The previous contractor had walked off after collecting ninety per cent of the money, so trust was already broken before we arrived. We earned it back the hard way.

But here is the thing. If I could have typed "what happens to emulsified fats and oils in a grease interceptor" into a machine and gotten an honest answer in thirty seconds, that entire year would not have happened. That is the tool clients have now, for free, and I would rather they use it on my proposals than find out the same way I did.

Five questions worth pasting along with the design

1. "Here is my measured wastewater characteristic and here is the design flow. Show me the calculation for the aeration tank volume." Ask for the arithmetic, not the verdict. Then compare it to the supplier's tank.
2. "Which numbers in my own documents contradict each other?"
3. "What does this design assume about my wastewater that nobody has measured?" The assumptions are where plants fail, not the equipment list.
4. "If this plant fails its discharge standard in year two, what is the single most likely reason?"
5. "What is not included in this scope?" Then take that list to the supplier and ask them to price it. That conversation tells you more about a company than any reference letter.

And one warning, from someone who uses these tools every working day. They will invent a number and present it with a straight face. They do not know what your local regulator will accept when the complaints start. They have never stood beside a plant at 2 in the morning and smelled that something was wrong before the lab results came back. Use it to check the reasoning. Do not let it sign the design.

A design that only survives because the client cannot check it is not a design. It is a sale.

If you want a second set of numbers to check a proposal against, our estimator will size and price a plant for your flow in a couple of minutes, and you can put that beside whatever you have been quoted.

If you have a proposal on your desk right now and something in it does not sit right with you, send it over. We are one of the few wastewater companies in the Philippines that will happily review a competitor's design with you, line by line, and tell you if it is fine. Book a free 15-minute consultation and bring the document.