Bahukhandi Labs is an independent software lab building AI products, developer tools and the research underneath them. A few things, made properly, and maintained for years — not a portfolio of demos.
Each product owns its problem end to end. They share a foundation, a design language and a standard — nothing else.
Legal research that shows its work.
NexLex reads statutes, judgments and filings, then answers questions with the passage it drew from attached to every claim. Built for people who cannot afford a confident wrong answer.
Reproducible dev environments that start instantly.
Learning that adapts to how far you actually got.
Anyone can claim their software is fast and private. The useful question is what they gave up for it. These are ours, with the cost attached.
Every interaction has a budget. Under 100ms feels like thought; over a second feels like waiting. We treat a regression in latency the same way we treat a regression in correctness — as a bug that blocks release.
A promise in a privacy page is worth very little. We prefer designs where the data never reaches us: local-first storage, per-tenant encryption, and defaults that collect nothing. What we cannot see, we cannot leak.
Care in an interface signals care underneath it. Typography, spacing and motion are not decoration — they are how software tells you whether it was made by someone paying attention.
Every feature we ship makes the next one harder. Most requests are better answered by making something existing work properly. Saying no is the main tool we have for keeping products understandable.
We build for the version of the product that exists in five years. That means boring dependencies, documented decisions, plain-file formats, and an export button that works before anyone asks for it.
Systems that assert without evidence are hard to trust and harder to debug. Where a product makes a claim — a citation, a score, a recommendation — it should be able to show you exactly where that came from.
Every product sits on the same foundation — one design language, one account, one standard. This is the whole lab, and it stays accurate because it is generated from the same registry the products are.
The shared foundation every product is built on. Written once, maintained once.
Tokens, typography and motion shared across every surface.
One account across the whole lab. One place to manage it.
Typed content in, versioned documentation out.
The products depend on problems nobody has fully solved. We write up what we find — including the results that do not flatter us.
An in-progress evaluation of scheduling review by topic mastery rather than per-item, using StudyFlow's early cohort data.
Grounded generation systems are routinely evaluated as if faithfulness to retrieved text were the same thing as being right. It is not. We separate the two and show where the confusion costs you.
A practical set of per-interaction time budgets, and what we give up to hold them. Measured on the products in this lab rather than borrowed from a 1968 paper.
A lab is a long game. Here is what has happened, what is happening now, and what we are aiming at — stated plainly enough that you can hold us to it.
2024
A run of small tools built to solve immediate problems. Nothing shipped publicly, but the habits formed here — plain files, fast startup, export by default — became the house style.
2025
Bahukhandi Labs is founded as an independent software lab. NexLex begins as an attempt to make legal research verifiable rather than merely fluent, and the grounding work behind it becomes the lab's technical centre of gravity.
2026
NexLex enters beta. Kiln and Groundwork spin out of internal infrastructure into their own projects. StudyFlow and Margin enter development. The shared platform — design system, auth, billing, docs — becomes the thing that makes each new product cheap to start.
2027
The goal is not more products. It is getting the current ones to the standard where people recommend them unprompted — general availability for NexLex, public releases for StudyFlow and Margin, and a published attribution benchmark.
Beyond
A small team, a research practice that publishes what it learns, and a portfolio of products that stay maintained. The measure of success is how much of this still works, and is still cared for, in ten years.
Whether you want to use one of our products, collaborate on research, or work here — the fastest route is a direct email.
hello@bahukhandi-labs.com