Dhwani

The company appears to be a prototype project named Dhwani, developed as part of the OpenAI 2026 hackathon. The description states that it is an acoustic biomarker screening tool for early childhood speech development assessment, intended for use by frontline workers in low-connectivity environments such as Anganwadi centers in India.

The project is self-reported and unverified — no revenue, customers, or traction data are provided. It is built with Flutter/Dart for the mobile app, Android native audio processing using Kotlin and ONNX models, and a Next.js dashboard for analytics. The system uses local processing to avoid sending raw audio to the cloud.

It implements a modest screening workflow that includes child profile creation, developmental questionnaires (in English and Malayalam), guided interaction activities, on-device acoustic analysis, and referral pathways via WhatsApp. The acoustic scoring is based on three biomarkers: VTTL, CVR, and PFV, with GREEN/YELLOW/RED tiers.

Key commercial due-diligence questions include:

• Whether the prototype has been validated in real-world settings.
• If there are plans to integrate questionnaire and acoustic signals.
• How the referral system will be localized and made secure.
• What clinical validation or regulatory compliance steps are planned.

The single most important open question is: Has this prototype been tested with actual frontline workers and children, and does it have any mechanism for ongoing clinical validation or integration into existing healthcare systems?

The description states that Dhwani is a mobile application designed to assess early childhood speech development using acoustic biomarkers derived from caregiver-child interactions.

It includes:

• A Flutter-based worker-facing app for Android and browser.
• On-device audio processing, including recording, voice activity detection (VAD), speaker segmentation via ONNX model, pitch tracking, and feature extraction.
• Acoustic scoring engine that evaluates three biomarkers: VTTL, CVR, and PFV.
• A local SQLite database for session storage.
• Optional Supabase sync for non-demo sessions.
• A Next.js dashboard for analytics (currently using mock data).
• A referral pathway via WhatsApp, with a PDF generator (not yet wired into the UI).

The app is designed to run offline and supports:

• Child profile creation
• Developmental goal review
• Parent questionnaire completion
• Guided interaction activities (Rattle, Toy hide & reveal, Imitate “aaa”)
• Local acoustic result display (GREEN/YELLOW/RED tier)
• Offline session saving and later sync

Not evidenced: No information on actual deployment, usage, or performance metrics.

The description states that Dhwani is a prototype tool aimed at helping frontline workers notice potential developmental concerns, particularly in early childhood (ages 12–36 months). It emphasizes:

• That it does not replace clinical assessment or caregiver judgment.
• That it focuses on non-semantic properties of interaction such as response timing, vocalization time, and pitch variation.
• That it uses on-device processing to maintain privacy and reduce reliance on connectivity.

The positioning is modest — the authors state that the goal is to provide a consistent additional signal, not a diagnostic tool. It is framed as a screening aid, not a replacement for validated tools or specialists.

Inference: The project appears to be positioned in the early-stage healthcare technology space, targeting underserved regions with limited access to specialists.

The description states that Dhwani is intended for frontline workers such as those in Anganwadi centers, which are part of India’s child welfare system. These workers:

• Often have limited time and unreliable connectivity
• Need tools that can help them identify when a screening may warrant closer attention
• Are expected to use the app during caregiver–child interactions

The app supports:

• Malayalam language for consent guidance
• Age-matched developmental questions (MyChild engine)
• Local session history and referral workflows

Not evidenced: No data on actual users, their training, or adoption rates. The ICP is inferred from the stated use case.

The description does not provide any information about:

• Revenue streams
• Pricing models
• Customer acquisition costs
• Monetization strategy

It only describes a prototype workflow for screening and referral, with no indication of how this would scale or be monetized in a commercial setting.

Not evidenced: No business model or pricing data.

The system is built using:

• Flutter/Dart for the mobile app
• Android native Kotlin for audio processing (AudioRecord, ONNX Runtime)
• ONNX model (Pyannote segmentation) bundled with SHA-256 verification
• SQLite for local data persistence
• Supabase for optional sync to cloud
• Next.js dashboard for analytics (mocked currently)

Key technical features include:

• On-device audio processing without sending raw PCM
• Local scoring engine using acoustic thresholds
• Evidence checks for recording quality
• Support for offline workflows with later sync

Inference: The prototype is built with a focus on local processing and minimal cloud dependency, which may be suitable for low-connectivity environments.

The description states that this is a hackathon submission (OpenAI 2026) and does not include any evidence of:

• Revenue
• Customers or users
• Product-market fit
• Adoption or usage data
• Clinical validation or regulatory approval

It is described as a prototype, with many features still in development or unimplemented, such as:

• Combining questionnaire and acoustic signals
• Full referral system localization (PDF generation not wired)
• Real-time sync retry/conflict handling

Not evidenced: No traction or maturity data.

The description does not mention any direct competitors. However, it implies that Dhwani is positioned in the early childhood development screening space, which may include:

• Traditional developmental screening tools
• Digital health platforms for child assessment
• AI-based speech and language development tools

It differs by focusing on acoustic biomarkers and on-device processing, which could be a unique value proposition if validated.

Not evidenced: No competitive landscape or market positioning data.

1. Prototype-only status: The system is described as a hackathon prototype with many unimplemented features.
2. No clinical validation: The description explicitly states that the acoustic scoring and age references are placeholders pending local clinical validation.
3. Incomplete referral system: PDF generation and localization (Malayalam) are not yet implemented.
4. Privacy and compliance risks: While privacy safeguards are described, no formal privacy impact assessment or ethics approval is mentioned.
5. Limited scalability: The app is designed for offline use but lacks robust sync retry logic or monitoring.
6. No monetization strategy: No indication of how the tool would be commercialized or scaled.

1. Has the prototype been tested with actual frontline workers and children in real-world settings?
2. What clinical validation steps are planned for the acoustic biomarkers (VTTL, CVR, PFV)?
3. How will the system handle uncertainty in audio quality or device performance?
4. Are there plans to integrate questionnaire and acoustic signals into a unified screening result?
5. What is the roadmap for localizing the referral system (PDF generation, WhatsApp message)?
6. Is there any plan for regulatory approval or ethical review before deployment?
7. How will the system be maintained and updated in the field?