Dhaka, working worldwide

A small studiothat would rathersay no.

We build software and AI systems for teams who have to defend the answer. Based in Dhaka, working with clients across time zones.

We work in Bangla and English at the same fidelity: Most language tooling is built and validated on English and then pointed at everything else. We start from the assumption that Bangla is a first-class input: its own tokenisation, its own OCR error profile, and code-switched text handled directly rather than routed by language detection.
Research methods, production delivery: Attribution, held-out evaluation and per-class reporting are not academic garnish; they are how you find out whether the thing works. We use them, and then we ship code that is typed, tested and in your repository from the first commit.
Explainability is a deliverable, not a slide: If a system makes a decision about a person, someone will eventually have to justify it. We build the evidence layer into the product rather than reconstructing it afterwards for a report.
We would rather say no: If a problem does not need machine learning, we will say so. A deterministic rule you can read beats a model you cannot explain to the person whose case was misrouted, and it does not need a bill per request.

Start here

You havesomething thathas to be right.

Tell us what it has to get right and what it is allowed to get wrong. If it is not something we should take on, we will say so and point you somewhere better.

Reply time: We reply within one working day.
Or write directly: zasif855@gmail.com
Where: Dhaka, Bangladesh