ASR model comparisonDivyajot SinghSep 28, 20266 min read
WER alone hides how ASR models fail. BRIDGE ASR 2.0 compares Gemini, GPT-4o, Saaras & Scribe on Indic, code-switched speech by error type.
Read Article→
body camera footage for aiPushpak AgrawalSep 10, 20264 min read
Body camera footage caps first-person AI training data at whatever hardware a program can issue. See why scaled, verified crowdsourcing gets egocentric data further.
Read Article→
WERDebashish GhoshSep 3, 20263 min read
A perfect transcript can still fail a WER test when code-switched words have two valid spellings. Here's how a normalization pipeline removes the scoring artifact.
Read Article→
what is supervised finetuningAnoushka AnanthAug 27, 20266 min read
Supervised fine-tuning is the stage that turns a pretrained VLA into a model that acts correctly. Here's how it works for robot foundation models.
Read Article→
vla modelBisma KaziAug 27, 20267 min read
VLA models need more than video. Here's exactly what kind of training data it takes to build a Vision-Language-Action model that actually generalizes.
Read Article→
what is vla modelArunabh RohatgiAug 27, 20265 min read
VLAs and LLMs share an architecture but not much else. Here's what actually separates a Vision-Language-Action model from a large language model.
Read Article→
open pi zeroDebashish GhoshAug 27, 20266 min read
Open pi 0 is the community effort to reproduce Physical Intelligence's pi 0 model in the open. Here's what it's built on and why data is the real bottleneck.
Read Article→
pi 0Pushpak AgrawalAug 24, 20266 min read
pi 0 (Pi-Zero) is Physical Intelligence's Vision-Language-Action model for general-purpose robots. Here's how it works and what it means for the field.
Read Article→
vlaBisma KaziAug 24, 20267 min read
VLAs let robots see, understand instructions, and act in real time. Learn how Vision-Language-Action models work and why real-world data drives them.
Read Article→
vlaAdnaan MohammedAug 24, 20266 min read
Get complete guidance on VLA training data with core modalities, data collection, annotation, datasets, and challenges while training Vision-Language-Action models.
Read Article→
ASR training dataDebashish GhoshJul 29, 20267 min read
Learn how voice and speech data for AI, accent diversity, and audio annotation play a crucial role to make AI models to understand real-world scenarios.
Read Article→
voice data collectionBisma KaziJul 29, 20268 min read
See how voice data collection actually works at scale - from verified contributor sourcing to validation, QC, and annotation for ASR, TTS, and speech AI models.
Read Article→