PanzaMail  by IST-DASLab

Email assistant trained and run on-device

Created 1 year ago
294 stars

Top 89.9% on SourcePulse

GitHubView on GitHub
Project Summary

Panza is a personal email assistant designed to mimic a user's writing style and provide contextually relevant email drafts. It targets individuals seeking an AI assistant that operates entirely on-device, ensuring data privacy and customization. The primary benefit is a personalized, secure email drafting experience without sharing sensitive communication data.

How It Works

Panza employs a two-stage process: data playback and local fine-tuning. First, it converts a user's exported emails (mbox format) into instruction-response pairs using a pre-trained LLM for summarization. This "data playback" technique trains the LLM to generate emails based on instructions, capturing the user's style. Second, it uses parameter-efficient fine-tuning methods like RoSA (combining LoRA and sparse finetuning) to adapt the LLM to the user's data locally. A Retrieval-Augmented Generation (RAG) module is then used during inference to provide relevant past emails as context for generating new drafts.

Quick Start & Requirements

Health Check
Last Commit

5 months ago

Responsiveness

Inactive

Pull Requests (30d)
0
Issues (30d)
0
Star History
2 stars in the last 30 days

Explore Similar Projects

Starred by Chip Huyen Chip Huyen(Author of "AI Engineering", "Designing Machine Learning Systems"), Pawel Garbacki Pawel Garbacki(Cofounder of Fireworks AI), and
4 more.

LongLoRA by dvlab-research

0.1%
3k
LongLoRA: Efficient fine-tuning for long-context LLMs
Created 2 years ago
Updated 1 year ago
Feedback? Help us improve.