<p>Last week, a Bengaluru-based techie posted on X about how Indian wedding buffets often leave guests spoilt for choice yet full of regret, with many piling their plates too soon or missing the best dishes. His solution, BuffetGPT, is an AI tool that scans a buffet and gives diners a “game plan” on what to eat, skip, and how much to consume. The post has since received 1.8 million views and 6,900 likes. </p>.<p>“It came purely out of frustration,” says 27-year-old Panathur resident Pankaj Tanwar, who got the idea at a friend’s haldi ceremony. “I built it in a day and tested it at the wedding reception the next day. It worked well,” he says. Tanwar tried around 13 to 14 dishes at the wedding, following the AI tool’s plan. “A funny outcome was when it recommended that I avoid the raita as it might fill me up. However, the spread included biryani, so it didn’t make sense to have it without raita,” he recalls.</p>.<p><strong>How it works</strong></p>.<p>The tool uses computer vision and pretrained models to detect dishes laid out at a buffet, through photographs uploaded to it. “When dishes are labelled, detection becomes easier,” he adds. Tanwar has primarily used the vision model YOLO v8 for object detection in the tool.</p>.<p class="bodytext">The tool essentially creates a personalised eating plan by factoring in personal data. “Users can add details such as their weight, daily calorie intake, and food preferences. Dishes are assigned a rank, which determines the order of consumption. Suitable quantities are also recommended,” he explains. Talking about technical challenges, he says that “generic large language model hallucination” (where output contains inaccuracies) occurs when trying to identify dishes from a regional cuisine it is not familiar with.</p>.<p class="CrossHead">What next?</p>.<p class="bodytext">Though it began as a side project, the online attention it received has pushed Tanwar to develop it further, he says. “Netizens are requesting features like calorie and macro tracking. Also, the concept could extend to restaurant spreads and food festivals,” he adds. </p>.<p class="bodytext">Despite the success of his post, it also received some flak, with an X user writing, “Thanks for finding a solution to a problem no one had.”</p>.<p class="bodytext">Currently in its alpha stage, BuffetGPT is expected to go online for public trials by next week. <span class="italic">For details, check @the2ndfloorguy on X.</span></p>
<p>Last week, a Bengaluru-based techie posted on X about how Indian wedding buffets often leave guests spoilt for choice yet full of regret, with many piling their plates too soon or missing the best dishes. His solution, BuffetGPT, is an AI tool that scans a buffet and gives diners a “game plan” on what to eat, skip, and how much to consume. The post has since received 1.8 million views and 6,900 likes. </p>.<p>“It came purely out of frustration,” says 27-year-old Panathur resident Pankaj Tanwar, who got the idea at a friend’s haldi ceremony. “I built it in a day and tested it at the wedding reception the next day. It worked well,” he says. Tanwar tried around 13 to 14 dishes at the wedding, following the AI tool’s plan. “A funny outcome was when it recommended that I avoid the raita as it might fill me up. However, the spread included biryani, so it didn’t make sense to have it without raita,” he recalls.</p>.<p><strong>How it works</strong></p>.<p>The tool uses computer vision and pretrained models to detect dishes laid out at a buffet, through photographs uploaded to it. “When dishes are labelled, detection becomes easier,” he adds. Tanwar has primarily used the vision model YOLO v8 for object detection in the tool.</p>.<p class="bodytext">The tool essentially creates a personalised eating plan by factoring in personal data. “Users can add details such as their weight, daily calorie intake, and food preferences. Dishes are assigned a rank, which determines the order of consumption. Suitable quantities are also recommended,” he explains. Talking about technical challenges, he says that “generic large language model hallucination” (where output contains inaccuracies) occurs when trying to identify dishes from a regional cuisine it is not familiar with.</p>.<p class="CrossHead">What next?</p>.<p class="bodytext">Though it began as a side project, the online attention it received has pushed Tanwar to develop it further, he says. “Netizens are requesting features like calorie and macro tracking. Also, the concept could extend to restaurant spreads and food festivals,” he adds. </p>.<p class="bodytext">Despite the success of his post, it also received some flak, with an X user writing, “Thanks for finding a solution to a problem no one had.”</p>.<p class="bodytext">Currently in its alpha stage, BuffetGPT is expected to go online for public trials by next week. <span class="italic">For details, check @the2ndfloorguy on X.</span></p>