The first thing you noticed about Bedrock, the machine-learning platform BasisAI shipped out of Singapore, was the audit trail. Not the model accuracy, not the deployment dashboard, but the lineage view: a quiet ledger showing which features pushed a credit decision one way or the other, who trained the model, when it last drifted. For a regulated bank in Southeast Asia in 2019, that ledger was the entire pitch. The math was table stakes. The receipt was the product.
BasisAI was founded in 2018 by Silvanus Lee, Liu Feng-Yuan, and Linus Lee, a trio TechCrunch described as Bay Area returnees coming home with Silicon Valley engineering credentials [TechCrunch, Jan 2019]. Liu, the CEO, had previously served as chief data scientist of the Singapore government [Asian Scientist Magazine], a posting that gave the company an unusually clear read on what regulators in the region were going to start asking of any enterprise deploying machine learning at scale. The bet was that explainable AI, often shortened to XAI, would move from academic concern to procurement requirement, and that whoever built the cleanest tooling for it in Asia would have a defensible wedge.
The bet
BasisAI sold to enterprises, not developers. Its core product, Bedrock, was positioned as a machine-learning platform built specifically around explainability and accountability rather than raw training throughput [AI Demand]. The company described its mission as building scalable and accountable machine-learning products [Vulcan Post], and it picked partners that reinforced the governance angle. In one of its more telling moves, BasisAI joined Open Loop as a dedicated private-sector technical assistance partner, helping enterprises develop AI explainability solutions on Bedrock [APNews PR Newswire]. Open Loop is an experimental governance program that pairs regulators with companies to draft workable AI policy. Showing up there is not a marketing exercise. It is a signal about which conversations the founders thought mattered.
The customer profile that fit was straightforward: a bank, an insurer, or a government-linked enterprise in Southeast Asia, sitting on a decade of data, being told by a board or a regulator that any model touching a customer decision needed to be inspectable. BasisAI's wedge was that it had built the inspection layer first, rather than bolting it onto a generic MLOps stack later.
Why the thesis had weight
The seed round told you who else believed it. In January 2019, BasisAI came out of stealth with $8.2 million from Sequoia Capital India and Temasek [DealStreetAsia][Tracxn Funding], one of the larger seed rounds out of Singapore at the time. Vulcan Post noted the same figure and the same investor pair [Vulcan Post]. That is a specific combination: Sequoia brings the venture-scale ambition, Temasek brings the institutional access to exactly the kind of regulated Southeast Asian buyer BasisAI needed in the room.
BasisAI Seed (Jan 2019) | 8.2 | $M
The tailwind was real and is now more obvious than it was then. Every major financial regulator in the region has since published guidance on AI fairness, transparency, and model risk. The Monetary Authority of Singapore's FEAT principles, released around the same window BasisAI was scaling, made explainability a first-class procurement question rather than a research topic. A startup with a government-trained CEO, a Singapore base, and a product designed around audit and lineage was sitting in roughly the right spot at roughly the right time.
The team and the trajectory
Liu Feng-Yuan led the company as CEO, with Silvanus Lee and Linus Lee as co-founders [Asian Scientist Magazine]. Around the founding team, operators like Tee Sing Tang and Juriani Tan were referenced as part of the early build [LinkedIn]. The combination, public-sector data leadership at the top, engineering returnees from the Bay Area, and operators with Singapore startup reps, was unusually well-matched to the buyer.
The traction story ended earlier than the funding pace suggested it might. In August 2021, roughly two and a half years after coming out of stealth, BasisAI was acquired by Aicadium [Crunchbase, Aug 2021], a Singapore-based AI company backed by Temasek. The team and the Bedrock platform folded into Aicadium's broader enterprise AI stack. For a seed-stage company with a deeptech product and a long enterprise sales cycle, an acquisition by a well-capitalized strategic inside its own investor's portfolio is a recognizable outcome, neither a runaway win nor a failure.
What bears would say, what bulls answer
The bear case on a company like BasisAI in 2020 and 2021 was straightforward: the enterprise MLOps category was getting crowded fast, with global platforms from the major clouds bundling explainability features into their existing model pipelines, which compresses the wedge for a standalone product even when the standalone product is better. The bull answer, supported by the company's Open Loop work [APNews PR Newswire] and the profile of its investors [DealStreetAsia], is that regulated buyers in Southeast Asia were not going to be served well by a generic US-cloud feature checkbox, and that a regional platform with regulator-fluent founders had a real claim to that buyer. The acquisition by Aicadium suggests the strategic value of the team and Bedrock was recognized; the question of whether the standalone business could have compounded into something larger is one the market did not get to fully answer.
What to watch
BasisAI as an independent company is now Aicadium's explainability and governance layer. The forward question for anyone tracking this thesis is whether the Bedrock lineage and audit tooling becomes a default inside Aicadium's enterprise deployments across Southeast Asia, and whether the founding team's regulator-facing posture continues to shape how AI gets sold into Singaporean and regional banks. The broader signal for founders watching from the outside is more durable: the company that got funded by Sequoia and Temasek out of Singapore in 2019 was not selling a model. It was selling the receipt.
Which raises the cultural question Bedrock was quietly answering all along: in a region where the regulator gets to the model before the customer does, is explainability the product, and is everything else just inference?