Features

1. Project Management

Manage documents and analysis results on a per-project basis.

Create / edit / delete projects
Language settings (Korean, English, Japanese)
Custom document analysis prompts
Per-project color themes

2. Document Upload & Preprocessing

Upload files in various formats and the system automatically detects the file type and routes them to the appropriate preprocessing pipeline. Supports documents (PDF, DOC, TXT), images (PNG, JPG, GIF, TIFF), videos (MP4, MOV, AVI), and audio files (MP3, WAV, FLAC) up to 500MB.

Drag-and-drop / multi-file upload
Automatic file type detection and pipeline routing
PaddleOCR (SageMaker) text extraction
Bedrock Data Automation document structure analysis (optional)
PDF text layer extraction (pypdf)
Audio/video transcription (AWS Transcribe)

For detailed preprocessing flows by file type, see Preprocessing Pipeline.

File Type	Supported Formats	Preprocessing
Documents	PDF, DOC, TXT	PaddleOCR + BDA (optional) + PDF text extraction
Images	PNG, JPG, GIF, TIFF	PaddleOCR + BDA (optional)
Videos	MP4, MOV, AVI	AWS Transcribe + BDA (optional)
Audio	MP3, WAV, FLAC	AWS Transcribe

3. AI Analysis Pipeline

Uploaded documents are automatically analyzed through a Step Functions workflow. A Strands SDK-based ReAct Agent performs in-depth analysis on each segment using an iterative question-answer approach.

Per-segment deep analysis (Claude Sonnet 4.6 Vision ReAct Agent)
Video analysis (TwelveLabs Pegasus 1.2)
Document summary generation (Claude Haiku 4.5)
Vector embedding and storage (Nova Embed 1024d → LanceDB)
Reanalysis / Q&A regeneration / Q&A add and delete

Document Upload
  → Preprocessing (OCR, BDA, Transcribe)
    → Segment Splitting
      → Distributed Map (max 30 concurrency)
        → ReAct Agent Analysis (per segment)
          → Vector Embedding → LanceDB Storage
      → Document Summary Generation

For detailed analysis flow, see AI Analysis Pipeline.

4. Real-time Notifications

Workflow progress is delivered to the frontend in real-time via WebSocket. DynamoDB Streams detects state changes, looks up active connections in Redis, and pushes events through the WebSocket API.

Step-level start / complete / error notifications
Segment analysis progress (X/Y completed)
Workflow start / complete / error notifications
Artifact and session creation events

5. Workflow Detail View

View detailed per-segment results for analyzed documents.

Per-segment OCR / BDA / PDF text / AI analysis results
Video segment timecode-based viewing
Transcription segment review
Q&A regeneration (with custom instructions)
Q&A add and delete
Full document reanalysis

6. AI Chat (Agent Core)

A conversational AI interface powered by Bedrock Agent Core. IDP Agent and Research Agent leverage tools via MCP Gateway for document search, artifact generation, and more, performing Q&A based on documents uploaded to the project.

Chat Features

Streaming responses with real-time tool usage display
Image/document attachments (multi-modal input)
Markdown rendering and code highlighting
Session management (create / rename / delete, conversation history preserved)

Hybrid Search

The AI Agent automatically searches project documents during conversations.

Vector search + Full-Text Search (FTS)
Korean morphological analysis (Kiwi)
Cohere Rerank v3.5 result reranking

Custom Agents

Create custom agents per project for specialized analysis.

Agent name and system prompt configuration
Create / edit / delete agents
Switch agents during conversations

MCP Tools

Tool	Description
search_documents	Hybrid search across project documents
save/load/edit_markdown	Create and edit markdown files
create_pdf, extract_pdf_text/tables	Generate PDFs and extract text/tables
create_docx, extract_docx_text/tables	Generate Word documents and extract text/tables
generate_image	AI image generation
code_interpreter	Python code execution

7. Artifact Management

Manage files generated by agents during AI chat (PDF, DOCX, images, markdown, etc.) in the artifact gallery.

Browse and search artifacts
Inline preview (images, PDF, markdown, HTML, DOCX)
Download and delete
Filter by project
Navigate to originating project