My first serious product project was SomnoAI Digital Sleep Lab. The starting idea was simple: if AI could understand more context and information, could it help make a complicated subject easier to explore?
I was not trying to build a broad technology company at the start. I was trying to make one project work. That turned out to be enough to expose almost every missing skill I needed next.
THE FIRST SHIFTFrom “I have an idea” to “how does this actually run?”
2025–2026 · LEARNING BY BUILDING
One product became a map of everything I did not know yet.
Building software pushed me far beyond AI itself. I had to learn about web interfaces, databases, authentication, cloud infrastructure, search, security, social systems and product design because each area became a real dependency.
01AI systemsModels, prompts, data and evaluation.
02Web productInterfaces, flows and responsive UX.
03DataDatabases, schemas and storage trade-offs.
04SearchCrawling, indexing and useful retrieval.
05CloudDeployment, limits and reliability.
06TrustSecurity, verification and privacy.
The pattern was consistent: I would build the first version, reality would expose a constraint, and the next version would have to be more deliberate.
THE DIFFICULT PARTS
The useful lessons arrived disguised as problems.
Code broke. Servers returned errors. API limits appeared at inconvenient moments. Compute and other resources were not always available. Applications were rejected. Some features worked in a demo but not well enough to keep. Others had to be rebuilt after a lot of time had already gone into them.
CODEDebug the evidence, not the ego.
Logs and reproducible failures are more useful than insisting the first idea was right.
INFRAConstraints are architecture inputs.
Rate limits, compute limits and deployment limits shape the product whether you like them or not.
PRODUCTRebuilding can be progress.
If the foundation is wrong, preserving sunk work can cost more than replacing it.
APPLICATIONSRejection is data, not identity.
A “no” can still reveal what the story, evidence or product needs next.
I would rather document the real process — including failed approaches and redesigns — than pretend every project worked on the first attempt.
2026 · THE EVOLUTION
SomnoAI stopped being the whole map.
As the work expanded, I realised the things I wanted to build were no longer limited to a single sleep-focused project. The broader direction became Vynalth AI.
The ecosystem widened into Vynalth AI Search, Vynalth AI Utility, Vynova and Vynalth AI Shield, alongside experiments around navigation and knowledge. Different products, one recurring question: can this become genuinely useful, understandable and reliable?
2025SomnoAI Digital Sleep LabFocused product exploration
→
2026Vynalth AIBroader AI company + product ecosystem
OUTSIDE THE PRODUCT · MCE 2026
Not every important milestone happened behind a screen.
MCE 2026 was a very different kind of team experience. The choir received 86.25 · Gold B in the Young Children (Open) category. The number matters, but so do the people, rehearsals, pressure and the ordinary final-day photo before going home.
TODAY · FROM MALAYSIA
Local roots. Global standards. No finished version.
I am still a student, and I am still learning. At the same time, I am building products publicly enough for real constraints to show up. That combination is the point: learning does not have to wait until everything feels official.
Malaysia is part of how I see technology — multilingual, multicultural, connected and constantly adapting. I want the work to carry that identity without making geography a ceiling.
Built from Malaysia. Open to the world.Dari Malaysia, untuk dunia.
THE CURRENT CHAPTER
Still learning. Still shipping.
The next chapter is not a slogan. It will be whatever gets built, tested, rejected, improved and eventually becomes useful enough to last.