IBM's Ben Hardill is describing what his colleague Bharat Bedi ate for dinner.
Hardill didn't share this meal or even catch the scent of cooking food. He wasn't even in the same house. All he needed to work out that Bedi had enjoyed a bowl of meatballs was a line on a graph.
That graph plotted how much energy was consumed by the Bedi household throughout the course of a day. So far, so boring. But this data, when analysed alongside information collected by other sensors in a home, can tell you a good amount about a person's daily habits, as you can see below.
Using the daily data gathered from the Bedi household IBM researchers were able to extrapolate what was taking place inside the house. The household's routine was worked out by analysing data charting the home's energy consumption, CO and CO2 levels, temperature and humidity throughout the day.
"For instance we taught the system to recognise when the dishwasher is on, this spike [in energy consumption] is a pattern that we know is the dishwasher, so we can see it's been on twice in a day," said Hardill.
After correlating patterns in the data the researchers were even able to take a stab at working out what Bedi was eating in the evenings, said Hardill, who is an emerging technologies and services consultant at IBM.
"It got a little bit more interesting in the evening as we started to be able to make guesses about what Bharat was having for dinner," he said.
"This spike here is the oven going on and the ripple afterwards is it keeping warm. Then there's a follow-up spike a little bit later, which happens to coincide with a huge change in humidity as well, but doesn't match up with the energy usage profile for the kettle.
"So we worked out that this was probably a pan of water going on for some pasta. So this was a meal of spaghetti and meatballs that night. We asked him a little bit later and he said 'I think we did have pasta and meatballs last week'."
The insights don't stop at what was on the menu. By rigging up homes with networked sensors IBM has been able to predict which rooms have people in and roughly how many people they contain (by changes in carbon dioxide levels), and whether someone has just lit up a cigarette (through changes in quantities of carbon monoxide).
The question of what can be done with such seemingly prosaic data as the amount of CO2 in a room, is becoming more important as smart energy meters and other small low-energy, networked sensors start to find their way into homes.
Google is betting that smart home technologies will be big business, with its recent $3.2bn acquisition of Nest Labs, the maker of programmable, networked thermostats and smoke and carbon monoxide detectors.
IBM's experiments with home sensor technology have found that data which can seem esoteric and unimportant can in aggregate reveal a lot of about what a person gets up to.
Watched over by machines of loving grace?
IBM has been experimenting with how these insights that sensor networks give you into daily routines could be put to good use. About four years ago the city of Bolzano in Italy was looking for a way to improve the care it provided to elderly residents in their homes without sending the cost of social care rocketing.
"Part of Bolzano's goal as a council is to become a place where the elderly of Italy go to retire in comfort, where they can live independent lives for as long as possible but still have the back-up they need," said Hardill.
To help achieve that goal the council teamed up with IBM's Human Centric Solutions team to fit 30 flats with networked smoke, carbon monoxide and dioxide, temperature, humidity and leaking water sensors. These sensors were connected to a GuruPlug fitted with a Zigbee wireless adapter that received data from the sensors and a 3G modem passed data back to IBM and the city council.
Data from these sensors was analysed and used to generate alerts that would be sent to council staff whose job it was to check on these residents, with messages relayed to Android phones carried by council staff.
"The alerts would go off so quickly that if there was a water leak and the warden was around he would come knocking on the door before water had even started to seep out from underneath the cupboards," said Hardill.
Where the system "started to get interesting", according to Hardill, is when IBM started to look for trends in the data from the sensors to allow them to infer what was happening inside the property.
"The carbon dioxide sensor is a classic one, if you watch the trend from that you can start to determine room occupancy for the space that it's in. If it starts to creep up you've probably got more people," he said.
By building patterns of everyday behaviour for each of the residents, IBM was even able to generate alerts for the council when an elderly resident deviated from their usual routine.
"If you let these systems learn and run for a little while you can build up [a picture of] systems of life," said Hardill.
"One of the things that fell out from Bolzano is you could learn that Mrs Smith, in number 25, normal day starts at about 8.30am and the first thing she does is put on her kettle and her toaster for breakfast and settles down with the newspaper.
"If by 9.30am you've not seen those spikes from the kettle and the toaster you can ask is there a problem? Should we be looking at the CO2 levels in her bedroom that day to see whether she's moved into the living room?
"We started to generate softer levels of alerts for the warden, just to get him to check that things were alright from time to time.
"It was a subtle alert, and you have to be careful because people do break pattern, for instance if the warden knows that Mrs Smith goes to spend some time with her daughter at the weekend then the warden knows not to intervene at this point.
"But those gentle pokes at the warden could allow them to be more attentive."
Two thirds of the elderly residents who took part in the pilot project reported an improvement in their quality of life and Bolanzo City Council reported a 31 percent saving to the cost of care for the elderly.
IBM is currently participating in a multi-million programme led by the University of Bristol to study how sensors systems can be used to develop a 24/7 digital home health assistant.The analytics applied to lives of the elderly residents in the Bolanzo study, and that may be applied more widely once such sensors find their way into more homes, have much in common with algorithms used to identify how computer systems are likely to fail, said Hardill, albeit with a few tweaks to deal with people rather than hardware.
"It's using the same sort of algorithms that are used for industrial predictive failure analysis. It's just softening the bounds a little bit because we're not talking about a hardened industrial system, we're talking about a human system."
Nick Heath is chief reporter for TechRepublic. He writes about the technology that IT decision makers need to know about, and the latest happenings in the European tech scene.