UpThink

[Upstage AI Ambassador] Personal Knowledge Management with Upstage Solar Pro 2 ✨

Overview

UpThink is a service designed to minimize the repetitive manual effort in Personal Knowledge Management environments (specifically Obsidian). It addresses the following bottlenecks that inevitably arise during the knowledge organization process.

Problem	Description
Image Data Processing	Manual conversion of visual information into text
Tag Management	Maintaining tag conventions and styling concerns
Lack of Knowledge Connectivity	Search costs for finding relevant past notes
Unstructured Documents	Need for splitting massive notes

UpThink automates these processes based on the powerful language understanding capabilities of Upstage Solar Pro 2. Users can break free from simple repetitive tasks and focus on what matters most—thinking.

Table of Contents

• Overview
• Key Features
• Tech Stack
• Architecture
  • Flow Chart
• Installation
• Usage
  • Watch Demo Video
• Project Structure
• Members & Roles
• Acknowledgements

Key Features

1️⃣ Image Alt Text Generation

Extracts text from images within notes and automatically generates alt text describing the image content.

• Perform OCR and extract document structure from images using Upstage Document Parse
• Generate alt text of around 50 words based on extracted text using Solar Pro 2
• Batch process all images within markdown files
• Automatically insert alt text below image links in ![[image.png]] format

2️⃣ Tag Recommendation

Analyzes note content to recommend appropriate tags and maintains consistency with existing Vault tag conventions.

• Automatically collect existing tags in Vault (supports hashtags #tag and YAML frontmatter)
• Set user-defined tag guidelines (language, case, separators, number of tags)
• Generate tags based on note content using Solar Pro 2
• Compare and match similarity with existing tags using Qwen Embedding model
• Automatically insert tags in YAML frontmatter format

3️⃣ Related Note Recommendation

Finds notes semantically similar to the current note and automatically connects them.

• Vectorize notes in Vault using Upstage Embedding Model and Chroma DB
• Automatically identify and batch process unembedded notes
• Process long notes through chunking
• Recommend Top 3 related notes via similarity search
• Automatically append backlinks to the ## Related Notes section using the [[note]] format

4️⃣ Note Splitting

Splits massive notes by topic to atomize them and build an interconnected knowledge system.

• Automatically extract topics within notes using Solar Pro 2
• Support flexible splitting strategies based on templates
• Edit, delete, or add extracted topics
• Automatically generate and save split atomic notes
• Automatically insert backlinks and ## Generated Atomic Notes section in original note

Tech Stack

Category	Technology
Language	Python 3.13
Frontend	Streamlit
LLM	Upstage Solar Pro 2
Document AI	Upstage Document Parse
Embedding	Upstage Embedding, Qwen3-Embedding-0.6B
Vector DB	Chroma DB
Framework	LangChain
Package Manager	uv

Architecture

Layer	Component	Description
Frontend	Streamlit	Web-based User Interface
Backend	Python Modules	Implementation of 4 core features
Upstage API	Solar Pro 2, Document Parse, Embedding Model	LLM, OCR, Vector Embedding
Local	Qwen Embedding, Chroma DB	Tag similarity comparison, Note vector storage

Flow Chart

Image Alt Text Generation

Step	Flow	Key Backend Modules
1	Extract image links and check alt text existence	MarkdownImageProcessor._collect_images_to_process()
1	Search image file paths in Vault	MarkdownImageProcessor._find_image_in_vault()
2	Extract text from image	OCRProcessor.extract_text()
2	Generate alt text	AltTextGenerator.generate_alt_text()
3	Insert alt text below image link	MarkdownImageProcessor.process_images()

Tag Recommendation

Step	Flow	Key Backend Modules
1	Collect and check existing tags	TagExtractor.get_unique_tags(), TagExtractor.count_tags()
2	Set tag guidelines and generate new tags	GuidelineGenerator(), TagGenerator.generate_tags()
3	Compare existing and new tags	TagComparator.compare_tags()
3	Suggest final tags	TagComparator.get_final_tags()
4	Insert YAML Frontmatter	add_yaml_frontmatter()

Related Note Recommendation

Step	Flow	Key Backend Modules
1	Identify unembedded notes	Related_Note.get_unembedded_notes()
2	Preprocessing and chunking	Related_Note.clean_text(), Related_Note.chunk_text()
2	Embed notes and save to DB	Related_Note.index_unembedded_notes()
3	Search related notes	Related_Note.find_related_notes()
4	Insert backlinks	Related_Note.append_related_links()

Note Splitting

Step	Flow	Key Backend Modules
1	Load prompt template	PromptLoader.load_template()
1	Extract Topic	UpstageClient.generate_with_template_sync()
2	Parse Topic list	ResponseParser.parse_topics_from_json()
3	Generate atomic notes	FileHandler.create_atomic_note()
3	Insert backlinks	FileHandler.insert_backlinks()

Installation

Supported Environments

• macOS
• Windows (PowerShell, CMD)

Install uv

• https://docs.astral.sh/uv/getting-started/installation/

Homebrew

brew install uv

Windows

powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Project Setup

# git clone
git clone https://github.com/geminii01/product-usecase-knowledge-management-upthink.git
cd product-usecase-knowledge-management-upthink

# Environment Variable Setup (Required!)
cp .env.example .env

# Open .env file and enter API keys
# UPSTAGE_API_KEY=your_api_key_here
# TAVILY_API_KEY=your_api_key_here

# Install Python 3.13 and dependencies automatically
uv sync

Run

streamlit run frontend/app.py

# Access via Local URL below!
# http://localhost:8501

Usage

🎬 Watch Demo Video: YouTube 🎬

Basic Usage

1. Enter Vault Path in the sidebar. (Absolute path to Obsidian Vault)
2. Upload Markdown file to process.
3. Go to the desired feature page and execute.

Project Structure

upthink/
├── frontend/                     # Streamlit Frontend
│   ├── app.py                    # Main App (Routing, Common Sidebar)
│   ├── home.py                   # Home Page
│   ├── image_ocr.py              # Image Alt Text Generation UI
│   ├── tag_suggest.py            # Tag Recommendation UI
│   ├── related_note.py           # Related Note Recommendation UI
│   ├── note_split.py             # Note Splitting UI
│   └── note_freshness.py         # Freshness Check UI
│
├── backend/                      # Backend Logic
│   ├── image_ocr/                # Image OCR & Alt Text Generation
│   │   ├── ocr_processor.py      # Document Parse API Integration
│   │   ├── alt_text_generator.py # Solar Pro 2 Alt Text Generation
│   │   └── markdown_processor.py # Markdown Image Processing
│   │
│   ├── tag_suggest/              # Tag Suggestion
│   │   ├── tag_extractor.py      # Extract 2 Tag Patterns
│   │   ├── tag_guidelines.py     # Guideline Generation
│   │   ├── tag_generator.py      # Solar Pro 2 Tag Generation
│   │   ├── tag_comparator.py     # Qwen Embedding Similarity Comparison
│   │   └── markdown_processor.py # YAML frontmatter Processing
│   │
│   ├── related_note/             # Related Note Recommendation
│   │   └── related_note.py       # Chroma DB based Similarity Search
│   │
│   ├── note_split/               # Note Splitting
│   │   ├── config.py             # Config
│   │   ├── models.py             # Data Models
│   │   ├── core/                 # State Management, File Handling
│   │   ├── llm/                  # LLM Client, Prompt Loader
│   │   └── ui/                   # UI Components
│   │
│   └── note_freshness/           # Freshness Verification
│       ├── api/                  # Tavily, Wikipedia API
│       ├── core/                 # State Management
│       └── llm/                  # LLM Integration
│
├── prompts/                      # Prompt Templates (YAML)
├── pyproject.toml                # Project Configuration & Dependencies
└── .env.example                  # Environment Variable Example

Members & Roles

Kim Su-yeon	Oh Ju-yeong	Yoon I-ji	Hong Jae-min
▪︎ Developed Image Alt Text Generation Feature	▪︎ Developed Note Splitting Feature ▪︎ Integrated Freshness Verification	▪︎ PM ▪︎ Developed Related Note Recommendation Feature	▪︎ Developed Tag Recommendation Feature ▪︎ GitHub Management & Team Code Integration

Acknowledgements

This project was conducted as part of the Upstage AI Ambassador program.
We thank Upstage for providing credits to support this project.