--- title: "Noteweave | AI Lab" description: "Noteweave Labs builds custom AI models, autonomous agents, document intelligence, and computer vision systems on your data — backed by published research in crowd counting, audio transformers, respiratory screening, and automated research critique." source: "https://labs.noteweave.io/" aeo_version: "1.0" --- # Noteweave Labs The applied AI lab at Noteweave. We turn proven research and your data into intelligent systems that ship fast and pay for themselves. Noteweave Labs builds custom AI models, autonomous agents, document intelligence, and computer vision systems — every engagement scoped to a clear business outcome, not a science project. Our work is grounded in peer-reviewed research published by the team. ## Site Map - [Labs Home](https://labs.noteweave.io/) — https://labs.noteweave.io/ - [Noteweave (main site)](https://noteweave.io/) — https://noteweave.io/ - [Web app (Log in)](https://apps.noteweave.io/) — https://apps.noteweave.io/ - [VS Code Extension](https://marketplace.visualstudio.com/items?itemName=noteweave.noteweave) — https://marketplace.visualstudio.com/items?itemName=noteweave.noteweave ## Services We turn proven research and your data into systems that ship fast and pay for themselves. Every engagement is scoped to a clear business outcome, not a science project. ### Custom AI models Models trained on your data and tuned to your use case. You get an edge competitors cannot copy, plus full ownership instead of a per-call API bill that grows with you. ### Autonomous agents Agents that run your repetitive, multi-step workflows end to end, inside the tools you already use. Cut hours of manual work and scale output without scaling headcount. ### Document intelligence Turn your contracts, reports, and PDFs into instant answers. Your team stops digging through files and gets cited results in seconds, so decisions move faster. ### Computer vision Put your existing cameras to work detecting objects, crowds, and safety risks in real time. No new hardware and no cloud fees, and your footage never leaves your premises. ## Projects Shipped prototypes and production pilots on client data. Screen recordings come from live pipelines, not mockups. ## Research Published work from the lab: crowd counting, audio transformers, respiratory screening, and automated research critique. [Google Scholar](https://scholar.google.com/citations?user=VV2eC_gAAAAJ&hl=en). ### Lightweight Feature Fusion for Crowd Counting (arXiv 2401.05968, Computer Vision, 2024) Crowd counting finds direct applications in real-world situations, making computational efficiency and performance crucial. Most previous methods rely on a heavy backbone and complex downstream architecture that restricts deployment. We introduce two lightweight models with MobileNet and MobileViT backbones, using Adjacent Feature Fusion to extract diverse scale features from a pre-trained model. [arxiv.org/abs/2401.05968](https://arxiv.org/abs/2401.05968) ### ASGIR: Audio Spectrogram-Guided Classification & Information Retrieval (arXiv 2407.18927) Recognition and interpretation of bird vocalizations are pivotal in ornithological research and ecological conservation. This paper presents ASGIR, an audio spectrogram-guided classification framework for improved bird sound recognition and information retrieval, with a two-step system that uses location and bird sounds to retrieve relevant information. [arxiv.org/abs/2407.18927](https://arxiv.org/abs/2407.18927) ### FGA: Fourier-Guided Attention for Crowd Counting (arXiv 2407.06110) Crowd counting is gaining societal relevance in urban planning, crowd management, and public safety. We introduce Fourier-guided attention (FGA), a novel mechanism that efficiently captures multi-scale global patterns via FFT, spatial attention, and channel-wise attention. FGA improves CSRNet and CANNet across ShanghaiTech, UCF-CC-50, and JHU++ benchmarks. [arxiv.org/abs/2407.06110](https://arxiv.org/abs/2407.06110) ### VoxMed: One-Step Respiratory Screening (arXiv 2407.18926) As respiratory illnesses become more common, quick and accurate detection is crucial for patient care. VoxMed is a UI-assisted one-step classifier using digital stethoscope recordings, with an Audio Spectrogram Transformer for feature extraction and a 1-D CNN to classify respiratory diseases in seconds. [arxiv.org/abs/2407.18926](https://arxiv.org/abs/2407.18926) ### E3: Automated Research Critique (arXiv 2605.27072, Research AI, 2026) We present E3, an automated review assistant that identifies decision-relevant technical concerns in research papers: unsupported claims, missing ablations, weak baselines, and threats to validity. To evaluate without contamination confounds, we adopt issue-level backtesting on papers postdating every model's training cutoff. [arxiv.org/abs/2605.27072](https://arxiv.org/abs/2605.27072) ## Team - [Yashwardhan Chaudhuri](https://www.linkedin.com/in/yashwardhan-chaudhuri) — Co-Founder, CEO - [Paridhi Mundra](https://www.linkedin.com/in/paridhi-mundra-1ba31622b) — Co-Founder, CPO - [Sanyam Jain](https://www.linkedin.com/in/sanyamjain2002) — Co-Founder, CTO ## Backed By - AWS Activate - Anthropic for Startups - xAI / Grok for Startups - Microsoft for Startups - OpenAI for Startups - Founders Inc ## Contact - Email: [contact@noteweave.io](https://mail.google.com/mail/?view=cm&fs=1&to=contact%40noteweave.io) - Book a 30-minute call: [cal.com/yashwardhanc/30min](https://app.cal.com/yashwardhanc/30min) - X / Twitter: [@noteweave_](https://x.com/noteweave_) ## Meta Copyright © 2026 Noteweave. All rights reserved. Turn operational problems into intelligent systems with us. Press `/` to toggle MACHINE mode. Press `Esc` to return to HUMAN.