A pilot pairing Singapore Institute of Technology students with Infocomm Media Development Authority mentors began placing multilingual ordering chatbots at selected hawker stalls this week, letting customers type or speak orders in Malay, Tamil, Mandarin or English before payment at the counter. The trial extends classroom prototypes such as FoodFriend—built for clearer stall communication—into live NEA-managed centres where queue pressure and language mix make mis-orders costly.
What the chatbots do on the floor
At participating stalls, QR codes open a lightweight chat that walks users through menu items with pictures, confirms spice levels, and repeats the order aloud in the language the customer chose. Stallholders still cash up in person; the bot’s job is to reduce pointing-and-guessing at peak lunch, not to replace hawkers’ judgment on sold-out dishes.
SIT teams said they trained models on anonymised order logs from earlier campus demos, then hardened prompts with IMDA’s Hawkers Go Digital vendors so personal data stays on-device where possible. Digital ambassadors from the SG Digital Office are on site during the first fortnight to help older patrons who prefer Tamil or Malay phrasing over English menus.
Why hawkers agreed to host
Owners who joined the pilot cited repeat tourists and elderly regulars who struggle with dialect-heavy specials boards. One chicken-rice stall in a central food centre reported fewer remakes after the bot began echoing orders in Tamil and Malay; another noodle stall said English-only delivery-app menus had been leaving money on the table for Malay-speaking shift workers nearby.
IMDA’s broader hawker digitalisation push—grants for e-payment and ordering—does not mandate chatbots, but the agency wants evidence that inclusive interfaces lift turnover without heavy commission fees. The SIT partnership keeps experimentation off commercial platforms’ template flows.
Students built separate language packs with stallholders vetting romanisation of dish names—char kway teow spelled three ways in old menus needed one canonical label before the bot could repeat orders accurately. Privacy officers reviewed logging so voice clips used for Tamil prompts are deleted after the session unless the patron opts into a feedback form.
Limits and what comes next
Chatbots stumble on code-switching and stall-specific slang; students monitor failure logs nightly. Halal trust still depends on human certification displayed at the counter—the bot repeats stall-declared labels but does not certify dishes. Payment remains PayNow or cash; no card data passes through student servers.
If error rates stay below manual mis-order baselines, IMDA may fold the stack into pre-approved Hawkers Go Digital bundles next year. Until then, the story is practical: multilingual bots at hawker stalls, built by SIT teams, tested where lunch queues already speak four languages.
