This is a submission for the Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle.
What I Built
Voice2Note is a local-first AI meeting follow-up assistant built for freelance developers, designers, and consultants.
Every client conversation can contain important decisions, project requirements, deadlines, and commitments. But after a meeting, freelancers still need to review their notes, identify action items, clarify responsibilities, and write follow-up emails.
Voice2Note turns that workflow into a structured process.
With Voice2Note, users can:
- Record or import voice recordings and generate transcripts.
- Extract meeting summaries, decisions, tasks, and client commitments.
- Distinguish confirmed deadlines from inferred or tentative dates.
- Identify unresolved client questions before starting implementation.
- Generate editable follow-up email drafts.
- Search across previous conversations using semantic search and grounded AI answers.
- Export tasks as CSV and meeting summaries as Markdown.
The application is designed to help freelancers reduce repetitive administrative work while retaining control over confidential client information.
Who would pay for it?
Independent developers, consultants, designers, and small agencies that regularly conduct client calls and need a reliable way to manage their follow-ups.
The business hypothesis is straightforward: if Voice2Note saves freelancers time after each meeting and helps them avoid missed commitments, a subscription could be worthwhile. That willingness to pay still needs validation with real users.
Demo
Live demo : https://voice2note-one.vercel.app
The demo should show a client-call recording being processed, the extracted tasks and commitments, deadline classification, a follow-up email draft, and the final export.
Code
GitHub repository: https://github.com/affanraza84/voice2note
The repository contains the application source code, setup instructions, technical documentation, and evaluation resources.
How I Built It
Voice2Note combines a Next.js application with local AI inference, structured extraction, and semantic retrieval.
Technology stack
- Next.js, React, TypeScript: Application interface and server-side functionality.
-
Whisper (
Xenova/whisper-tiny.en): Local speech-to-text transcription using Transformers.js and ONNX. - Llama 3.2 3B via Ollama: Local language-model inference for structured extraction and meeting intelligence.
- all-MiniLM-L6-v2: Local text embeddings for semantic search.
- SQLite: Embedded storage for application data and vector embeddings.
- Zod: Validation of structured AI outputs.
The processing pipeline transforms audio into a transcript, extracts structured information, and indexes the content for future retrieval.
One important design decision is to preserve evidence from the transcript alongside extracted tasks and commitments. Users should be able to verify what the AI found instead of blindly trusting a generated summary.
Voice2Note also distinguishes between explicit dates and inferred timeframes. A tentative statement such as “probably sometime next week” should not silently become a confirmed deadline.
The project includes privacy-oriented processing and storage configuration. The local-private workflow is intended to keep recordings and derived information on the user's machine; this guarantee should be verified against the actual runtime configuration before relying on it for confidential client material.
Why Does Open Innovation Matter?
Open innovation is central to Voice2Note because privacy, cost, and control are important for independent professionals.
1. Greater control over confidential information
Local inference makes it possible to process recordings without sending their contents to a hosted AI inference API, provided the local-only configuration is correctly enforced and verified.
2. Lower marginal inference costs
Once the required models are installed, local inference can avoid per-minute transcription and per-token inference charges. Hardware, electricity, setup, and maintenance still have costs, so local processing is not literally cost-free.
3. Model flexibility
Open-weight models and modular provider interfaces let developers evaluate alternative models, adjust prompts, and choose inference runtimes based on their hardware and requirements.
4. Transparency and reproducibility
The model choices, application code, processing pipeline, and evaluation approach can be inspected and tested. This gives developers more control than relying entirely on a closed hosted service.
For this project, open innovation is not just a technology preference. It is part of the product's value proposition: helping freelancers manage their client conversations without automatically making a third-party AI provider the custodian of that information.
Top comments (0)