Projects.

HeartVoice began as an internal initiative at Omron Healthcare — an attempt to spin a data‑provider model out of a hardware‑first company. I led the engineering build‑out and launch as the flagship Asia‑Pacific expansion, handling the full technology lifecycle: securing investment, assembling the engineering team, defining the architecture, and overseeing delivery from concept through commercial launch.

The platform integrates device data streams from Omron medical‑grade wearables with a cloud‑native backend, serving telemedicine providers and B2B health partners across APAC via API. Building it inside a regulated environment meant leading the team through ISO 13485 and broader regulatory workstreams, alongside ASEAN data‑privacy frameworks across six countries and the organisational inertia of a 90‑year‑old Japanese manufacturer. It passed all compliance review and launched commercially.

I continue as Strategic Technology Advisor — providing architectural guidance, technology roadmapping, and hands‑on direction to the engineering team.

One of the patterns that runs through my career is prototype building across very different environments: measurement systems for scientists, experimental platforms for clinical researchers, and decision‑support prototypes for business teams trying to validate a new market. The common thread is translation. Someone understands the domain deeply, but the idea is still too early, too ambiguous, or too cross‑functional to hand directly to a standard delivery team.

At A*STAR and in Japan, that meant working close to researchers on biosensors, microfabrication, optogenetic stimulation, and measurement systems where hardware behaviour and software instrumentation had to be understood together. At Duke‑NUS, it meant turning scientific questions into usable R&D workflows for neuroscience teams. At Omron Healthcare, the same skill matured into building prototypes that combined device data, cloud software, operational assumptions, and commercial analysis so business development teams could judge whether an opportunity deserved investment.

I am at my best in that early stage: when the problem crosses hardware, software, and business logic; when the stakeholders do not speak the same technical language yet; and when the fastest way to reduce uncertainty is to build a credible prototype with enough analytical depth to support a real decision.

I use my own VPS infrastructure as an R&D environment, deploying and running production‑grade AI agents to explore what autonomous workflows actually look like outside controlled demos. Five agents are running live, each serving a different function, all orchestrated via systemd services and Telegram‑based delivery.

Job hunting agent: Node.js multi‑source job fetcher pulling from Adzuna, The Muse, and Indeed RSS; roles scored for fit by Claude Sonnet; filtered results delivered to Telegram. Runs nightly at midnight SGT via systemd timer.

Financial briefing agent: Self‑executing daily market intelligence prompt covering equities (VOO, VTI, AAPL, MSFT, GOOG, AMZN, ORCL, MU, SMCI, MRK), commodities (WTI, Brent, spot gold), and ETFs (IAU, XLE). Delivered in Bengali — news and sentiment‑focused rather than price‑scraping.

OpenClaw autonomous agent: Configured with Slack integration, Telegram bot delivery, LinkedIn Playwright monitoring skill, and a news‑fetching automation layer. Runs on Ubuntu VPS with full exec elevation and Python‑based approval workflows.

Singapore Bus Telegram bot: On‑demand bus arrival times for home and office stops, responding to stop codes using the LTA DataMall API. Deployed as a systemd service.

AWS IaC pipeline: Claude Code + Terraform + GitHub Actions CI/CD for managing AWS infrastructure remotely — EC2, VPC, Nginx, S3 state backend, DynamoDB state locking, targeting ap‑southeast‑1.