We build AI agents for the CX teams that couldn't afford to keep hiring their way out of the problem.
Harry Yu spent five years as VP of CX Operations at one of Singapore's largest telcos. The operation looked fine on paper — 120 agents, 8,000 tickets a day, robust escalation flows. The problem was that CSAT was stuck at 62% and nobody could explain why.
He pulled three months of call recordings and read through 2,000 escalated tickets himself. The pattern was clear: customers weren't angry about the problem. They were angry about having to explain it four times to four different people. Every handoff reset the context. Agents were spending 40% of each call reading prior notes instead of solving anything.
The first prototype ran on a laptop in a conference room in 2022. It handled 80 real tickets in a weekend without a human in the loop. By end of 2022 it was in production, handling 35% of inbound tickets. CSAT went from 62% to 79% in six months — not because the product got better, but because customers stopped repeating themselves.
Level3 AI spun out as an independent company in early 2023. Sovereigns Capital led a $13M Seed Round in late 2024 to expand the platform across Southeast Asia.
Sovereigns Capital is a faith-driven venture fund focused on early-stage technology companies and lower-middle-market businesses with strong unit economics and genuine market need. They led Level3 AI's $13M Seed Round in 2024, bringing patient capital and a network of enterprise relationships across Southeast Asia and the US. The investment is being used to expand the engineering team in Singapore and build out sales coverage across Indonesia, Thailand, and the Philippines.
Every AI action your agent takes should be explainable to your ops team. We log the reasoning chain behind every decision — not a black box result. If the agent issues a refund, you can see why.
Explainability isn't a feature. It's how your team builds trust with the AI and deploys it further over time.
We don't try to automate everything. An AI that attempts to handle an angry customer's insurance dispute by itself will make it worse. Level3 AI is opinionated about escalation — it passes to humans earlier, not later, when stakes are high.
The goal isn't a zero-agent team. It's a smaller, better-targeted human team spending time on cases that actually need judgment.
Most conversational AI is built in English-first, Western-market environments and then poorly adapted for APAC. We built for APAC from day one. That means code-switching, regional slang, language-agnostic sentiment detection, and compliance with local data laws.
It also means pricing and packaging that reflects APAC enterprise buying cycles, not Silicon Valley SaaS assumptions.
We do not use customer conversations to train shared models. We do not sell usage data to third parties. Your tenancy is isolated. Your data can be exported or deleted on request within 72 hours.
We're building for enterprise trust over the long term. Short-term data exploitation would destroy that. It's not a close call.
Harry Yu and Mei Lin build an internal tool handling live tickets at a Singapore telco. First 80 conversations processed without human intervention.
Tool handles 35% of telco's inbound tickets. CSAT climbs from 62% to 74% within the first four months.
Company spins out as an independent entity. First external customer — a Singaporean e-commerce platform — goes live in 60 days.
Platform expands to 12 APAC languages. First Indonesian and Thai enterprise customers deploy in-region.
Sovereigns Capital leads $13M Seed Round. Engineering team doubles. Sales coverage expands to Philippines and Vietnam.
Book a call with Harry or Rajan. No pitch deck, no sales cycle — just a conversation about your CX challenges.
Get in Touch Meet the Team