The advent of Generative AI promises to bring about a productivity revolution. But what does that mean in practice? This ChatGPT case study offers a practical example of both its language processing and coding capabilities. 

When I was 17, I had already tried my hand at a few jobs. A paper round, bussing tables at the John Lewis coffee shop, and working at my local deli on Saturdays. I considered myself to be relatively industrious. But I would never have contemplated delivering $50k worth of value in a year; let alone a few weeks. That’s exactly what Harry Allison (aged 17) did on his internship. Harry is the son of my former boss Chris Allison, and I had the pleasure of speaking to them a couple of weeks ago. Here is how this unusual ChatGPT case study goes:

The Challenge

Harry started his internship with My Food Source, a company that specialises in food provenance analytics . To complement its main data sources, the company wanted to scan relevant news (national and international) to detect incidents of food contamination – outbreaks of salmonella, listeria, e.coli, botulism etc.

The information was relatively simple to access. All 33 news sources (spread across three continents) provided RSS feeds. The challenge was how to trawl through all that content and identify stories relating to outbreaks of food contamination.

To accomplish such a task in days of yore, an intern would have waded through stacks of newspapers each day. But times have changed. Harry sought to build a fully automated solution. And in achieving that goal, ChatGPT proved to be an incredibly useful companion.

A new era for language processing

Harry began using traditional open-source sentiment libraries like NLTK, VADER and TextBlob (python modules that calculate sentiment using legacy natural language processing techniques) to analyse the text from the RSS feeds. The idea was to identify outbreaks of food contamination by using a combination of key word search (for salmonella, listeria etc.) and sentiment scores (filtering out positive news like ‘company X adopts new technology to prevent e.coli’). But after validating the output of initial test runs, he found the model was only achieving 50% accuracy.

Running the same content through ChatGPT 4, Harry found the accuracy leapt to 95%. Not only was he able to identify the right news stories more accurately, but the new solution was much simpler. He no longer needed to specify keywords and instead could rely on ChatGPT’s semantic understanding to find any stories relating to ‘food borne illnesses.’

In addition to identifying the right articles, Harry was also able to ask ChatGPT to provide a summary of each. This made the end product – a daily newsletter summarising outbreaks of food borne illnesses – much more user friendly (readers no longer had to click through to the full story to get information beyond the headline).

ChatGPT’s multilingual capabilities also made it easier to expand coverage. Food chains are global, and My Food Source wanted to track news in several languages. This could be accomplished with very little overhead using ChatGPT, which works in up to 50 languages.

These improvements (improved accuracy, simpler specs, richer functionality, and multilingual coverage) cannot be seen as incremental; they are game changing. They make the difference between an automated process that requires a significant amount of manual intervention for each run; and an automated process that can be left to run autonomously (and just reviewed periodically).

The art of coding, without coding

Given that he’s only 17, it won’t surprise you to know that Harry is a relative novice when it comes to coding. To prepare himself for his internship, he had taken a two-day foundation course in Python, which taught him how to code ‘for loops’ and ‘if/then’ statements. But beyond that, ChatGPT provided all the help Harry needed.

In response to his prompts, ChatGPT helped write code snippets that would upload his chosen content to the gpt API and draw in his desired output. Once he had developed a functional block of code (to upload all the RSS feeds and aggregate the desired outputs), Harry even asked ChatGPT if it could help him optimize that code, which it duly obliged with.

Gauging the net impact

At the end of his internship, Harry achieved what he had set out to accomplish. He had developed an automated script that would aggregate dozens of global news sources, pick out articles relating to outbreaks of food contamination, and generate a newsletter that summarised text from each of those articles. This is now being used daily by My Food Source to spot outbreaks of food contamination from Toronto to Buenos Aires.

Some kids his age would have felt productive helping to tidy the stationary cupboard. But Harry had created something of lasting value. How much value? About $50,000. That is what his father estimates he would have quoted for a similar project only a few years ago. This can’t be discarded simply as the boast of a proud father. It is the educated opinion of someone with decades of industry experience; the founder of a highly successful IT consultancy who has delivered dozens of similar engagements.

A New ChatGPT Case Study: Follow Harry for a daily roundup of GenAI news!

Having developed his solution for My Food Source, Harry has repurposed the same code to generate a daily roundup of news relating to generative AI. He now publishes that via LinkedIn. Of course, ChatGPT also helped him write the script to fully automate his posts by publishing via the LinkedIn API. So, if you’re interested in staying up to date with the latest developments in GenAI – give Harry Allison a follow on LinkedIn!

ChatGPT CAse Study: Harry Allison