Observed arrival · 2026-10-13
Bodham Labs tests AI on the phones people own—and in the languages they speak
An independent Delhi research lab documents work on on-device speech, English and Hinglish models, agent memory, and AI for learning.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- AI researchers and builders working on on-device speech and Indian-language models
- Worth noticing
- The voice-memory comparison reports 11.7% and 24.5% English word-error rates across 17 of the lab’s own recordings.
Field notes
The homepage sorts its work by research line and marks entries as measured, exploring, or prototype, making the status of individual projects legible. A visual replay explains a hybrid design in which a state-space model runs an always-on loop, external memory stores hard facts, and a transformer handles harder decisions. The lab says it ultimately dropped state-space memory from a voice-memo app because its fixed-size memory tradeoff was a poor fit for short texts.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
ƒJavaScriptBrowser-side code central
One card from the complete issue