A customer support inbox usually looks like a backlog: questions to answer, complaints to resolve, and product details to clarify before the next message arrives.

For Aaryan Kandiah, it looked like something else: a live stream of business signals.

 That shift helped him win the AI Workflow Competition at Echelon Singapore 2026 with SignalDesk. A recent Nanyang Technological University graduate with a Bachelor of Engineering in Electrical and Electronics Engineering, Aaryan is also set to begin a Master of Computing in AI at the National University of Singapore. His winning workflow, built around Boldr’s customer support challenge, reflected that blend of engineering discipline and applied AI thinking.

 The point is not simply speed. If the AI cannot answer with evidence, it should flag the gap so the company can improve its knowledge base, FAQs, product pages, or internal documentation.

 If a question could not be answered, the system should not guess. It should flag a knowledge gap. Those patterns should become useful business signals.

Built from real SME friction

The competition was built around a practical premise: builders should work on real SME bottlenecks, not imagined use cases built for a stage demo. Over a 48-hour worksprint, participants were asked to build functional AI workflows for business problems faced by participating SMEs.

For Boldr, a Singapore-based watch micro-brand, the problem sat inside customer support. Like many small teams, Boldr deals with repeated enquiries across product information, policies, specifications, and purchase-related concerns. Together, these messages reveal what customers do not understand and where support teams lose time repeating answers.

 SignalDesk treats that inbox not as a queue to be cleared, but as a signal feed that can help the business learn. The workflow ingests a customer enquiry, identifies the likely intent, checks approved sources, and determines whether there is enough evidence to support a reply. If there is, it drafts a response for human approval. If there is not, it records the issue as a knowledge gap.

Also Read Inside the AI Workflow Competition at Echelon Singapore 2026

 That matters because customer support is not simply a language-generation problem. For an SME, one unsupported AI-generated reply can create confusion, damage trust, or create more work later.

Why it was not just another chatbot

Aaryan said the competition’s brief made it clear that builders were expected to go beyond a prompt-based chatbot.

That instruction made it clear that the expected outcome was a production-ready tool that is more complex than a detailed system prompt.

 SignalDesk’s most practical design choice is restraint. The workflow does not assume that every question deserves an automated answer. It first checks whether the business has enough verified information. If not, it stops short of responding and pushes the missing information back to the team.

That makes the human-in-the-loop layer central rather than decorative. A support agent still approves customer-facing replies, resolves missing information, and reviews suggested updates before publication.

In other words, SignalDesk does not remove human judgement from the process. It moves people away from repetitive first-draft work and towards decisions that require accountability.

The e27 layer behind the build

The workflow did not emerge in a vacuum. Before the worksprint began, the e27 team had turned SME pain points into structured challenge tracks, issued a Builders Kit, set submission requirements, and created official communication channels for announcements, questions, and peer support.

Builders were given two broad tracks. Revenue Rocket focused on sales, marketing, and customer acquisition, while Save-a-Hire focused on operational efficiency and task automation. Boldr’s challenge sat naturally within Revenue Rocket because repeated support questions can expose revenue leaks: unclear product information, weak customer education, or unanswered concerns that stop buyers from moving forward.

The competition also gave builders a clear operating frame: sponsor workshops with FPT AI Factory, Qwen, and Bitdeer AI, a virtual kick-off ceremony, and the timed release of official problem statements and sample materials on Day 1. Submissions had to show a working demo, business impact, cost analysis, safeguards, and proof of execution.

That structure shaped the kind of solution that could win. SignalDesk was not rewarded merely for generating a neat answer. Its evidence checks, human approval queue, and knowledge-gap logging matched a judging lens that looked attechnical execution, SME value, cost realism, responsible AI, and clarity.

What the winner left with

The win gave Aaryan more than stage recognition. He left with more than US$16,000 worth of prizes, credits, and post-competition support intended to help continue the winning workflow beyond the event.

The package included an e27 editorial feature to tell the SignalDesk story across Southeast Asia and exclusive SME matchmaking with businesses looking for practical AI workflows. It also included a 3-month Notion Business Plan, valued at US$6,000 in workspace credits, to support documentation, workflow planning, and collaboration.

On the technical side, the package included US$1,000 in Bitdeer AI compute credits, US$500 in Alibaba Qwen cloud and AI credits, PixVerse credits worth 400 minutes of generated video for demos and product storytelling, and US$6,000 in AMD-based cloud credits to test and scale AI workflows.

For SignalDesk, those resources matter because the project does not have to end as a competition prototype. Editorial visibility can explain the workflow to a wider market, SME matchmaking can open commercial conversations, and the credits can support further testing, refinement, demonstrations, and deployment exploration.

From inbox to operating system

The broader lesson is not that every SME needs an AI support bot. It is that many SMEs already sit on operational data they are not using well.

SignalDesk shows one way to make that shift. It starts with a familiar pain point, adds evidence checks and human review, and turns unanswered questions into a system for organisational learning.

That is why the winning workflow fits the spirit of the competition. It does not treat AI as spectacle. It treats AI as infrastructure for a business problem that already exists. For Southeast Asian SMEs, useful AI stories may begin not with a model, but with unresolved work waiting to be understood.

=====

Want updates like this delivered directly? Join our WhatsApp channel and stay in the loop.

The e27 team produced this article

We can share your story at e27 too! Engage the Southeast Asian tech ecosystem by bringing your story to the world.You can reach out to us here to get started.