SYNORA ZENITH Start a project

Home/Research

Work we do without a client attached.

We keep roughly a day a week for internal problems. Mostly these are gaps we hit repeatedly in client work and got tired of working around.

Current focus

Four open threads.

Status below is honest. "Active" means someone is on it this month. "Early" means we have written notes and a prototype, nothing more. "Paused" means it is parked until there is a reason to pick it up.

Devanagari text handling on the web

Nepali text on the web still breaks in predictable places: conjunct rendering across font fallbacks, incorrect line breaking, search and sorting that ignores Unicode normalisation, and form validation written for Latin script that silently rejects valid names. We are building an internal test suite and a set of fixes we can apply to every bilingual project instead of rediscovering the same bugs.

Unicode Typography Localisation
Active

Performance budgets for slow connections

Almost every performance guide assumes a fast connection and a recent phone. We are collecting field measurements from mid-range Android devices on mobile data around Kathmandu and outside the valley, and turning them into a budget we hold ourselves to: a target for first paint, total transfer size and time to interactive under those conditions rather than lab conditions.

Web performance Field data Applied on this site
Active

Inference at the edge

Running small models close to the user instead of round-tripping to a distant data centre. The interesting question for us is not whether it works, but where the crossover point sits: at what model size and request volume does edge inference actually beat a hosted API on both latency and cost, for the traffic patterns our clients have. We are benchmarking rather than assuming.

Edge compute Latency Cost modelling
Early

Offline-first field data collection

Health posts, schools and survey teams routinely work where there is no usable connection. The requirement is not "works offline" but "syncs correctly afterwards" — conflict resolution, duplicate handling, and knowing which record won and why. We are testing local storage and sync strategies against deliberately bad networks.

Sync Data integrity Field tools
Early

Why we publish this

We would rather be judged on how we think.

A portfolio shows finished screens. It does not tell you whether the team understood the problem or got lucky. Writing up what we are still working out is a more honest signal, and it means the next client with the same problem starts further along than we did.

If any of the above overlaps with something you are dealing with, we are happy to talk about it whether or not there is a project in it.

Get in touch