Skip to content

vnknowledge2014/file_translated_app

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
backend
backend
 
 
data
data
 
 
docs
docs
 
 
frontend
frontend
 
 
scripts
scripts
 
 
.dockerignore
.dockerignore
 
 
.env.example
.env.example
 
 
.gitignore
.gitignore
 
 
Dockerfile
Dockerfile
 
 
README.md
README.md
 
 
docker-compose.yml
docker-compose.yml
 
 

Repository files navigation

InfiTrans Logo

InfiTrans

Enterprise Document Translation Platform
by Infinite Agent Lab

Multi-Model LLM Routing · Solana Web3 Auth · 15+ Languages · Format-Preserving

Overview

InfiTrans is an enterprise-grade document translation platform designed for security, accuracy, and scale. Initially built for air-gapped local LLMs, InfiTrans now features a Multi-Model Routing Engine capable of dispatching translation workloads to OpenRouter, OpenAI, Anthropic, Gemini, or local Ollama instances based on language pair and domain.

Key Highlights

  • Multi-Model LLM Routing — Dynamically route tasks (e.g., ja->vi to Claude 3.5, and en->zh to GPT-4o) via OpenAI-compatible endpoints.
  • Enterprise Billing & Web3 Auth — Integrated Solana Pay for seamless, decentralized enterprise billing and wallet-based authentication.
  • Air-Gapped Ready — Fully supports 100% local inference via Ollama for zero-trust environments.
  • 15+ Languages & 7 Domains — Auto-detect source + translate to 15+ languages across Legal, IT, Medical, Finance, and more.
  • Format-Preserving — Translates DOCX, XLSX, PPTX, PDF, CSV, TXT, MD while preserving 100% of the original formatting.
  • XLIFF Exchange — Dual-version (1.2 + 2.1) export/import for CAT tool workflows.
  • Human-in-the-Loop — Confidence scoring + in-app bilingual review editor.

System Architecture

graph TB
    subgraph Client ["Browser (SvelteKit SPA)"]
        UI[InfiTrans UI<br/>Web3 Wallet Integration]
    end

    subgraph Docker ["Docker Compose Stack"]
        subgraph App ["FastAPI Application :8000"]
            API[REST API<br/>JWT / Solana Auth]
            ORCH[Orchestrator<br/>Pipeline]
            WP[Worker Pool<br/>Async Queue]
        end
        
        subgraph Storage ["Persistent Storage"]
            SURREAL[(SurrealDB<br/>v2.1.4)]
            MINIO[(MinIO<br/>Object Storage)]
        end
    end
    
    subgraph LLM ["LLM Engine"]
        ROUTER[Model Router]
        OLLAMA[Local Ollama]
        CLOUD[Cloud Providers<br/>OpenAI, Anthropic, etc.]
    end

    UI -->|HTTPS + JWT / Web3| API
    API --> WP
    WP --> ORCH
    ORCH --> ROUTER
    ROUTER -->|Local Inference| OLLAMA
    ROUTER -->|OpenAI-Compat API| CLOUD
    ORCH -->|Read/Write| MINIO
    API -->|CRUD| SURREAL
    ORCH -->|Job Status| SURREAL
Loading

Translation Pipeline (5 Phases)

graph LR
    A[📄 Upload] --> B[Extract]
    B --> C[Translate]
    C --> D[Score]
    D --> E[Review]
    E --> F[Reconstruct]
    F --> G[📥 Download]
    
    B -- "zipfile + xml.etree<br/>Inline Tags" --> C
    C -- "Ollama LLM<br/>Batch + Cache" --> D
    D -- "Confidence<br/>HIGH/MED/LOW" --> E
    E -- "Human Edit<br/>or Auto-Approve" --> F
    F -- "ZIP Clone<br/>Tag Restore" --> G
Loading
Phase Engine What It Does
Extract zipfile + xml.etree Walk XML trees, build inline-tag strings <tagX> → segments[]
Translate Multi-Model Router Route to best LLM (Local/Cloud) based on MODEL_{SRC}_{TGT} rules
Score Multi-signal heuristic Confidence scoring (HIGH/MEDIUM/LOW) per segment
Review Web Editor / XLIFF In-app bilingual editing or CAT tool roundtrip
Reconstruct Deterministic Python ZIP clone → replace text → format-preserved output

Supported Languages (15)

Language Code Language Code
Japanese ja Thai th
Vietnamese vi Indonesian id
Chinese (Simplified) zh Russian ru
Chinese (Traditional) zh-TW Portuguese pt
Korean ko Arabic ar
English en Hindi hi
French fr — —
German de — —
Spanish es — —

Source language supports Auto Detect mode using Unicode block analysis (CJK, Hiragana, Katakana, Hangul, Arabic, Devanagari, Thai, Cyrillic).

Supported Formats

Format Extraction Strategy Reconstruction Strategy
DOCX word/document.xml + headers/footers/endnotes <w:p> aggregation Non-destructive ZIP clone + inline tag restore
XLSX xl/sharedStrings.xml + worksheets + drawings + sheet names Byte-level surgery: sheet rename, formula ref update, calcChain drop
PPTX ppt/slides/slide*.xml <a:p> aggregation Non-destructive ZIP clone + inline tag restore
TXT/MD Line-by-line + diagram token extraction Line replacement + ASCII art grid expansion
CSV Cell-by-cell scan Cell replacement
PDF Text extraction (read-only) (reconstruction not yet implemented)

Quick Start

Prerequisites

  • Docker + Docker Compose
  • Ollama (with model pulled)
  • 8GB+ RAM (for 1.8B model) or 16GB+ (for larger models)

1. Configure Environment

cp .env.example .env
# Edit .env — defaults work for most setups

2. Pull Translation Model

ollama pull gemma4:31b-cloud

3. Start Services

docker compose up -d --build

4. Access UI

Open http://localhost:8000

Default credentials:

  • Username: infitrans_test
  • Password: InfiTest2026!

5. CLI Usage (Alternative)

cd backend && pip install -r requirements.txt

# Basic translation
python scripts/cli.py translate --file doc.docx
# Specify language pair
python scripts/cli.py translate --file doc.docx --source ja --target vi
# Export XLIFF for external review
python scripts/cli.py translate --file doc.docx --export-xliff --xliff-version 2.1
# Interactive terminal review
python scripts/cli.py review data/output/doc_vi.xlf
# Reconstruct document from XLIFF
python scripts/cli.py translate --file doc.docx --import-xliff data/output/doc_vi.xlf

Configuration

All settings via environment variables. Priority: env vars > .env file > defaults.

Core Settings

Variable Default Description
OLLAMA_URL http://ollama:11434 Ollama API endpoint
OLLAMA_TIMEOUT 1800 Request timeout (seconds)
MODEL gemma4:31b-cloud Translation LLM model
SOURCE_LANG auto Default source language
TARGET_LANG en Default target language
DEFAULT_DOMAIN general Default translation domain

LLM Inference

Variable Default Description
TRANSLATION_TEMPERATURE 0.7 Sampling temperature
TOP_K 20 Top-K sampling
TOP_P 0.6 Nucleus sampling
REPETITION_PENALTY 1.05 Repetition penalty
TRANSLATION_NUM_CTX 4096 Context window size
OLLAMA_THINK false Enable CoT reasoning mode

Extraction & Batching

Variable Default Description
MAX_INLINE_TAGS 8 Max tags before stripping
MAX_SEGMENT_CHARS 400 Max chars per segment
BATCH_MAX_CHARS 3000 Max chars per LLM batch
BATCH_MAX_SEGMENTS 5 Max segments per batch
MAX_CONCURRENT_BATCHES 2 Parallel batches

Security

Variable Default Description
SECRET_KEY (must change) JWT signing secret
STORAGE_ENCRYPTION_KEY (optional) SSE-C encryption key (base64, requires HTTPS)
FILE_RETENTION_DAYS 14 Auto-delete files after N days

API Endpoints

Method Endpoint Auth Description
POST /api/auth/register — Register new user
POST /api/auth/login — Login (returns JWT)
POST /api/upload JWT Upload document for translation
GET /api/jobs JWT List recent jobs
GET /api/jobs/{id} JWT Get job detail
GET /api/download/{id} JWT Download translated file
GET /api/download/{id}?xliff=true JWT Download XLIFF
POST /api/glossary/upload JWT Upload CSV glossary
GET /api/glossary JWT List glossary terms
DELETE /api/glossary/{id} JWT Delete term
GET /api/languages — List supported languages
GET /api/domains — List supported domains
GET /api/health — Health check

Project Structure

infitrans/
├── frontend/                     # SvelteKit SPA + Paraglide i18n
│   ├── src/
│   │   ├── lib/
│   │   │   ├── components/           # UI components (7 Svelte + SVG icons)
│   │   │   ├── stores/               # Reactive state (config, i18n)
│   │   │   └── api.ts                # Centralized API client
│   │   └── routes/
│   │       ├── (marketing)/          # Landing page (public)
│   │       ├── (auth)/               # Login / Register / Forgot Password
│   │       └── (app)/                # Translate workspace (protected)
│   └── messages/                     # i18n JSON (en, vi, ja)
├── backend/
│   ├── app/
│   │   ├── agent/                    # Translation pipeline core
│   │   │   ├── orchestrator.py           # Pipeline coordinator
│   │   │   ├── extractor.py              # Text extraction (Native OOXML)
│   │   │   ├── translator.py             # LLM batch translation + cache
│   │   │   ├── reconstructor/            # Format-preserving reconstruction
│   │   │   │   ├── docx.py / xlsx.py / pptx.py / plaintext.py
│   │   │   │   ├── _ooxml.py             # Shared OOXML utilities
│   │   │   │   └── _common.py            # Translation map builder
│   │   │   ├── confidence.py             # Confidence scoring
│   │   │   ├── prompt_router.py          # Domain/language prompt router
│   │   │   └── xliff.py                  # XLIFF 1.2/2.1 export/import
│   │   ├── llm/                      # LLM abstraction layer
│   │   │   ├── base.py                   # Abstract LLMClient
│   │   │   ├── factory.py                # Backend factory (Ollama/Cloud)
│   │   │   ├── model_router.py           # Tiered multi-model resolution
│   │   │   └── openai_compat.py          # Universal OpenAI API client
│   │   ├── ollama/                   # Ollama HTTP client
│   │   ├── prompts/                  # Prompt engineering resources
│   │   │   ├── rules/formats/            # OOXML / plaintext rules
│   │   │   └── skills/
│   │   │       ├── languages/source/     # 15 source language prompts
│   │   │       ├── languages/target/     # 15 target language prompts
│   │   │       └── domains/              # 7 domain prompts
│   │   ├── routes/                   # Admin, Billing, Wallet Auth, Jobs
│   │   ├── utils/                    # Language detection, file detect
│   │   ├── auth.py                   # JWT + Wallet Auth
│   │   ├── config.py                 # Environment settings
│   │   ├── database.py               # SurrealDB async client
│   │   ├── storage.py                # MinIO object storage
│   │   └── worker.py                 # Async job worker pool
│   └── tests/                        # Unit + integration tests
├── docs/                             # Architecture maps & guidelines
├── scripts/                          # Utility scripts
│   ├── cli.py                        # Unified CLI tool
│   └── i18n_manager.py               # Frontend translation sync
├── docker-compose.yml                # Docker deployment
├── Dockerfile                        # Multi-stage build (Node + Python)
└── .env.example                      # Configuration template

Translation Quality Controls

Control Implementation
Language-aware prompts Per-source/target language prompt rules with linguistic guidance
Domain-aware prompts Per-domain terminology and style rules
Inline Tag preservation Trados-style <tagX> serialization in LLM system prompt
Tag Validator (RALPH Loop) Regex catches missing/hallucinated tags, triggers automated retry
Source Leak Detector Unicode regex detects untranslated source text, triggers 1-by-1 retry
Count Mismatch Fallback Auto-retries 1-by-1 if batch response has wrong segment count
Translation Cache SurrealDB cache — clean translations skip LLM on subsequent encounters
Glossary Enforcement User-defined terms injected as mandatory translation table
Confidence Scoring Multi-signal heuristic → HIGH/MEDIUM/LOW triage
In-App Review Web bilingual editor + CLI for editing before reconstruction
XLIFF Roundtrip Edit in Trados/memoQ → re-import → reconstruct

Tech Stack

Layer Technology
Frontend SvelteKit 2 + Paraglide-JS (i18n) + TypeScript
Backend FastAPI + Python 3.13 + asyncio
LLM Ollama + gemma4:31b-cloud (31B params)
Database SurrealDB v2.1.4
Object Storage MinIO (S3-compatible)
Auth JWT + bcrypt (native, no passlib)
Deployment Docker Compose (multi-stage build)

License

Dual-licensing strategy:

1. Open Source (Free): Licensed under GNU AGPL v3.0. You may use, learn from, and modify this code. If you distribute or provide it as a network service, you must release your entire product's source code under AGPL v3.0.

2. Commercial (Paid): For closed-source commercial use without AGPL obligations, purchase a Commercial License.

Contact: office@infiniteagentlabs.com

Infinite Agent Lab · Building the future of local AI

About

No description, website, or topics provided.

Resources

Readme
Activity
Custom properties

Stars

0 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages