Add DMP '25 Week 12 blog: context-aware Arabic translation pipeline

Author: ac-mmi

src/constants/MarkdownFiles/posts/dmp-25-AmanChadha-week12.md

@@ -1,93 +1,71 @@
 ---
-title: "DMP '25 Final Report by Aman Chadha"
-excerpt: "Concluding my DMP '25 project: Migrating Music Blocks’ localization from webL10n.js to i18next and building an AI-assisted translation system with contextual support."
+title: "DMP '25 Weekly Update: Context-Aware AI Translation Pipeline"
+excerpt: "Final week progress: Generating full context data, testing the AI-assisted translation pipeline, and producing Arabic translations using Google Translate API."
 category: "DEVELOPER NEWS"
 date: "2025-09-08"
-slug: "2025-09-08-dmp-25-aman-chadha-final"
+slug: "2025-09-08-dmp-25-weekly-update-aman-chadha"
 author: "@/constants/MarkdownFiles/authors/aman-chadha.md"
-tags: "dmp25,sugarlabs,finalreport,aman-chadha"
+tags: "dmp25,sugarlabs,weeklyupdate,aman-chadha"
 image: "assets/Images/c4gt_DMP.png"
 ---
 
-<!-- markdownlint-disable -->
-
-# Final Report by Aman Chadha
+# Weekly Update: AI-Assisted Translation Pipeline Progress
 
 **Project:** [JS Internationalization with AI Translation Support](https://github.com/sugarlabs/musicblocks/pull/4731)  
 **Mentors:** [Walter Bender](https://github.com/walterbender), [Devin Ulibarri](https://github.com/devinulibarri)  
-**Duration:** July – September 2025  
+**Week:** September 8 – September 14, 2025  
 
 ---
 
-## Project Overview
+## Full Context Extraction Completed
 
-The aim of this project was to modernize the **internationalization (i18n) system** of Music Blocks by migrating from the legacy `webL10n.js` framework to **i18next**, and to introduce an **AI-assisted translation workflow** to reduce the burden on human translators.  
+This week, I finalized the **generation of the full context file** for Music Blocks UI strings. Using my **RAG (Retrieval-Augmented Generation) model**, I extracted:  
+- All `msgid`s from `.po` files.  
+- Surrounding 5 lines of code above and below each string.  
+- Any developer comments related to the string.  
 
-Key goals included:  
-- Making the i18n workflow **cleaner, modular, and maintainable**.  
-- Automating missing translations with **AI + context-awareness**.  
-- Supporting community-driven refinements of translations.  
+All extracted snippets were consolidated into a single **JSON file** with metadata including: source file, line numbers, and code snippets. This file is now ready for semantic indexing in **ChromaDB**, enabling fast and context-aware retrieval during translation.
 
 ---
 
-## Achievements
-
-1. **Migration from webL10n.js to i18next**  
-   - Replaced outdated framework with i18next for modern i18n support.  
-   - Added flexible fallback strategies (cleaned text, lowercase, title case, hyphenated).  
-   - Enabled JSON-based translation files for better maintainability.  
-
-2. **AI-Assisted Translation System**  
-   - Designed a pipeline to parse `.po` files and extract `msgid`s.  
-   - Generated **context automatically** by analyzing where each string occurs in the codebase.  
-   - Built a **RAG (Retrieval-Augmented Generation) model** to store these contexts.  
-   - Integrated **Google Translate API** to auto-fill missing translations using this context.  
+## Testing the AI-Assisted Translation Pipeline
 
-3. **Contributor-Friendly Workflow**  
-   - Human translators can review AI-generated suggestions instead of starting from scratch.  
-   - New language files can be created or updated automatically.  
-   - Significantly lowers the barrier for contributors to help Music Blocks reach more learners.  
+With the context data ready, I started testing **language translator models**:  
+- Generated explanations of UI strings using **Ollama’s language model**.  
+- Integrated **Google Translate API** to auto-fill missing translations, producing the first **Arabic `.po` file** for Music Blocks.  
 
----
+### Why Google Translate?
 
-## Challenges & Solutions
+While evaluating translation options, I considered open-source alternatives like **LibreTranslate**:  
+- **LibreTranslate / Argos Translate**: free and open-source.  
+- **Drawback:** Translates word-by-word without understanding the surrounding code context, which can cause ambiguous translations for strings with multiple meanings (e.g., “duck” for pitch vs. volume).  
 
-- **Challenge:** Extracting meaningful context for each translation key.  
-  **Solution:** Implemented a RAG approach that links `msgid`s to their source code usage.  
+**Google Translate API** was chosen because:  
+- Handles context better in practical UI usage.  
+- Integrates seamlessly with Python scripts and the RAG pipeline.  
+- Produces more accurate and human-readable translations for the Arabic `.po` file.  
 
-- **Challenge:** Ensuring migration didn’t break existing functionality.  
-  **Solution:** Incremental testing with sample `.json` files and stepwise replacement of webL10n.  
+The pipeline is designed to be **pluggable**, so other translation APIs like DeepL or OpenAI can be added in the future.
 
 ---
 
-## Key Learnings
+## Outcome This Week
 
-- Infrastructure-level improvements (like i18n) may not be flashy, but they **unlock global accessibility**.  
-- Context is essential for high-quality translations — raw machine translation alone isn’t enough.  
-- Clean migration strategy + thorough testing makes adoption smoother for the community.  
+- Successfully generated **full context JSON** for all UI strings.  
+- Tested retrieval and translation pipeline with Arabic.  
+- Generated a **working Arabic `.po` file** with contextual translations.  
+- Verified that translations respect the intended meaning of each UI string, reducing ambiguity for end-users.  
 
 ---
 
-## Future Work
+## Next Steps
 
-- Add support for more AI translation providers (e.g., DeepL, OpenAI).  
-- Build a simple **web-based review UI** for translators to accept/refine AI suggestions.  
-- Automate detection of new/changed strings via GitHub Actions and update translation files dynamically.  
+- Complete remaining `.po` files for additional languages (Japanese, Hindi).  
+- Integrate translation workflow with the Music Blocks repository via **PRs**.  
+- Build a simple **web-based review UI** for translators to refine AI suggestions.  
 
 ---
 
-## Closing Thoughts
-
-This project was a unique opportunity to contribute infrastructure that strengthens Music Blocks for a **global user base**. By combining **modern i18n practices with AI translation support**, I hope this work helps learners worldwide access Music Blocks in their own languages with greater ease.  
-
-I’m deeply grateful to my mentors Walter and Devin, and to the Sugar Labs community, for their support and guidance throughout this journey.  
+## Reflection
 
----
-
-## Resources & References
-
-- **Music Blocks Repository:** [github.com/sugarlabs/musicblocks](https://github.com/sugarlabs/musicblocks)  
-- **GitHub PR:** [#4731](https://github.com/sugarlabs/musicblocks/pull/4731)  
-- **i18next Documentation:** [i18next.com](https://www.i18next.com/)  
-
----
+This week was crucial in validating the RAG-based pipeline and demonstrating the **value of context-aware AI translations**. Choosing the right translation model and generating comprehensive context sets the foundation for scaling Music Blocks’ localization to many more languages, increasing accessibility and engagement for learners worldwide.