Become a TechRadar Insider
A better option than build or buy
Running a major event means juggling logistics, partnerships, programming, audiences and budgets – often simultaneously and at pace. The last thing you need is a technology project that creates more problems than it solves, which is the tension most event organisers face when it comes to apps and digital tools. Build something custom and you're looking at significant cost, long development cycles and ongoing maintenance. Pick something off the shelf and you're squeezing your content into someone else's framework.
When the Melbourne International Comedy Festival (MICF) wanted to help audiences navigate nearly 700 shows, neither option quite fit. The festival needed something that could understand the nuances of comedy – including the edgy stuff – respond to open-ended questions like a knowledgeable friend and point people in the right direction. That meant AI.
AI in a public-facing, content-rich environment brought real considerations: compliance, privacy, content safety and guardrails – and the ongoing commitment to get those things right with each iteration. Datacom partnered with MICF to build Funny Finder – an AI-powered personal entertainment concierge designed with robust guardrails from day one and refined continuously based on real audience behaviour. By taking on the technical complexity, Datacom enabled the festival team to focus on what they do best – while audiences got a seamless path from "what should I see?" to a planned and booked night out.
Discovery fatigue at scale
MICF's programme is vast and diverse. Scrolling hundreds of listings hoping something catches your eye quickly goes from discovery to fatigue. Audiences were burning out before they'd even bought a ticket.
Standard event apps couldn't solve this. Basic tagging and filtering can't meaningfully categorise comedy – a genre defined by tone, delivery, energy and language that often deliberately pushes boundaries. The festival needed something that could parse natural-language questions, handle group preferences and recommend a full night out – safely and reliably.
A concierge, not a catalogue
Funny Finder works as a conversational concierge embedded in the MICF website. Festivalgoers describe what they like – or don't – and the tool responds with clear, personalised recommendations drawn only from verified festival data.
In 2025, it attracted up to 6,000 daily interactions, which became the foundation for improvement in 2026. This year, new features let users get artist-matched recommendations from an increased programme of nearly 800 shows, blend group preferences and build a full evening plan with dining suggestions. Every enhancement was shaped by observed user behaviour and developed in close collaboration with the MICF team.
These efforts led to an uptick in usage, with up to 12,000 daily interactions in 2026. More than 240,000 requests were made by over 40,000 people across upwards of 7,000 chats during this year's festival period.
Guardrails that evolve
Comedy is intentionally provocative, which put Funny Finder on a direct collision course with standard AI content safety filters. Show descriptions sometimes tripped blocking rules. Getting this right required – and continues to require – deliberate, layered work.
Datacom treated the model as untrusted until proven otherwise and implemented a two-pronged guardrail strategy: Azure's content filters remained in place for user prompts and outputs, while official show descriptions were pre-processed to preserve creative intent without triggering false positives. When the model started riffing during testing, the team dialled it back to concierge mode – eliminating an entire class of risk.
Both teams stress-tested the tool with unexpected, off-the-wall questions to tighten responses and build confidence in the framework. Guardrails aren't set-and-forget – Datacom monitors, tests and refines with each iteration, working alongside the MICF team.
Built for control, designed for iteration
Funny Finder was built on Microsoft Foundry using Azure's platform-as-a-service infrastructure, with GPT-family models as the core language engine. The environment is hosted within a scalable Azure tenancy on Datacom's own cloud, with rate-limiting controls to ensure uptime during traffic surges.
Rather than fine-tuning a static model on comedy content, Datacom implemented a pattern similar to retrieval-augmented generation (RAG). A custom pipeline pulls live data from the festival's CMS into a vector database, ensuring only vetted, current content reaches the model. Every answer is checked against the source dataset before it reaches the user. Security measures including API gateways, web application firewalls and rate-limiting guard against prompt injection and distributed denial-of-service (DDoS) attacks.
Your event, the same rigorous approach
The technology behind Funny Finder was designed to be replicated and tailored – different content, different audiences, same rigorous approach to guardrails, content safety and ongoing refinement.
"When money and funding is tight, trying new opportunities can feel impossible. But that's where partnerships thrive," says MICF Marketing and Partnership Director Denise Damianos. "Datacom brought the know-how and a motivated, talented team. Innovation doesn't happen when you play it safe. It happens when you find the right partner, trust the process and have the courage to take the leap."
"A great festival night should feel effortless. Funny Finder now helps people shape the kind of night out they want – without juggling multiple screens or tabs."
"We learned a lot about how people used the tool last year and that has informed the adaptations," says Datacom Director of AI, Lou Compagnone. "The new features keep Funny Finder simple, helpful and human – and they show what's possible when AI design is grounded in genuine collaboration and user insights."
If you're running an event and want to give your audience a smarter, safer way to engage with your programme, talk to Datacom about how a tailored AI concierge could work for you.