RecipeVault.Ai

Ankit Sharma — Sun, 04 Oct 2026 19:12:45 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

My grandfather has spent decades perfecting his culinary skills, documenting his process through dozens of rambling, beautiful audio memos. However, these recordings aren't just lists of ingredients; they are woven with obscure regional dialect terms, family history, and anecdotes about where he sourced his spices. Standard speech-to-text tools completely butcher his dialect, and generic AI models strip away the charm of his stories when trying to extract the recipes.

To solve this, I built RecipeVault AI, an offline-first, local AI tool designed specifically for him. It takes his raw audio recordings, accurately transcribes his unique dialect, and intelligently separates the nostalgic family lore from the actual cooking steps. The end result is a beautifully formatted, structured Markdown recipe book that retains his personal touch without getting bogged down in transcription errors.

When I showed him the final product, his reaction was priceless. He was absolutely amazed that it correctly captured his highly guarded "secret sauce" recipe—including the exact pinch of regional spices—without uploading a single byte of his voice to the cloud!

Demo

Watch the RecipeVault AI Demo on YouTube
(Alternatively, try the live demo here: https://recipevault-demo-placeholder.com)

Code

ankitsharma-14 / RecipeVault.Ai

A 100% offline, privacy-first AI tool that uses Whisper and local Llama 3 to transcribe family audio memos, decode regional dialects, and separate cherished family lore from structured cooking instructions into Markdown recipe books

🍲 RecipeVault AI

RecipeVault AI is a 100% offline, privacy-first application built to preserve family history. It transcribes local audio recordings containing obscure regional dialects and family stories, extracts structured cooking instructions, and generates a formatted Markdown recipe book.

Built for the Hacktoberfest Weekend Challenge: Build for a Friend.

Tech Stack

Frontend: Streamlit
Audio Transcription: Whisper (OpenAI open weights)
LLM: Llama 3 (via Ollama)
Orchestration: LangChain

Why Local AI?

This project relies entirely on local, open-source AI to guarantee complete privacy for personal family stories. It works 100% offline, incurs zero API costs, and avoids the risk of cloud speech models misinterpreting highly specific regional dialects.

Quick Start

Prerequisites

Install Ollama and pull the Llama 3 model:
```
ollama run llama3
```

View on GitHub

How I Built It

Building this required a fully local pipeline to ensure zero latency and maximum privacy. Here is a breakdown of the architecture:

Frontend UI: I used Streamlit to create a simple, drag-and-drop interface for my grandfather. It's clean, intuitive, and runs perfectly on his local machine.

Transcription: For audio processing, I integrated Whisper (OpenAI's open weights). Running the base or small model locally was powerful enough to pick up his regional slang far better than cloud-based alternatives, especially when provided with a custom initial prompt.

LLM & Orchestration: The heavy lifting for text extraction is handled by Ollama running Llama 3 locally, orchestrated via LangChain.

The Pipeline: Once Whisper generates the raw transcript, LangChain passes it to Llama 3 with a highly specific prompt template. The LLM is instructed to identify and structure the ingredients and steps into a Markdown format, while summarizing the "family lore" into a neat introductory paragraph for each recipe

Why Does Open Innovation Matter?

For a project this personal, open-source AI wasn't just a preference; it was a strict requirement. Using closed APIs like OpenAI or Anthropic would have meant sending highly personal family stories and voices to third-party servers.

Open innovation allowed RecipeVault AI to work 100% offline, guaranteeing absolute privacy for our family's history. Furthermore, running local open-weight models means there is zero API cost, ensuring my grandfather can process as many hours of audio as he wants without worrying about a subscription bill. The ability to heavily customize local prompt templates in LangChain also meant I could fine-tune the system to understand his specific regional dialect in a way rigid cloud APIs couldn't support

DEV Community: Ankit Sharma