screenpipe
screenpipe copied to clipboard
[bounty] script to fine tune local LLM or OpenAI on your screenpipe data
/bounty 100
definition of done:
- simple to use script (can be python, whatever) to fine tune model (LLM like llama3.2 or multimodal or OpenAI) on your screenpipe data
- some docs to run it and then use the LLM in your pipes or screenpipe app settings (like ollama or openai)
💎 $100 bounty • mediar.ai
Steps to solve:
-
Start working: Comment
/attempt #717with your implementation plan -
Submit work: Create a pull request including
/claim #717in the PR body to claim the bounty - Receive payment: 100% of the bounty is received 2-5 days post-reward. Make sure you are eligible for payouts
❗ Important guidelines:
- To claim a bounty, you need to provide a short demo video of your changes in your pull request
- If anything is unclear, ask for clarification before starting as this will help avoid potential rework
- Low quality AI PRs will not receive review and will be closed
- Do not ask to be assigned unless you've contributed before
Thank you for contributing to mediar-ai/screenpipe!
| Attempt | Started (UTC) | Solution | Actions |
|---|---|---|---|
| 🟢 @mfenderov | Jan 15, 2026, 01:53:10 PM | #2047 | Reward |
| 🟢 @ | Feb 17, 2025, 11:52:58 AM | WIP | |
| 🟢 @Quanta-Naut | Sep 20, 2025, 09:44:43 AM | WIP |
/attempt #717
/attempt #717
Implementation Plan
I'll create a Python script to fine-tune LLMs on screenpipe data with the following approach:
Deliverables
-
finetune_screenpipe.py- Main script supporting:- Local fine-tuning with Llama 3.2 (via unsloth/transformers)
- OpenAI fine-tuning API option
- Automatic data extraction from screenpipe SQLite DB
-
Data Processing
- Extract OCR text, audio transcripts from screenpipe DB
- Format as instruction-following dataset (Q&A pairs)
- Support conversation-style and completion-style formats
-
Documentation
- Clear README with installation steps
- Examples for both local (ollama) and cloud (OpenAI) workflows
- Integration guide for screenpipe pipes/settings
Timeline
- Day 1: Data extraction + formatting
- Day 2: Fine-tuning scripts + docs
- Day 3: Testing + PR submission