Application Changelog

Model Selection Locked: The AI model is now fully auto-selected by workflow rules and is no longer user-editable in the UI.

Whisper Removed: Removed OpenAI Whisper from the UI and runtime paths. The toolbox now uses Gemini models only.

Inference Policy Update: Added task-specific generation settings with high thinking enabled for both paths. Transcription uses temperature 0.1; translation uses temperature 0.3. Runs are auto-forced to gemini-3.1-pro-preview when high-thinking policy is active.

Prompt Consistency: Added a strict transcription-first instruction prefix at the beginning of auto-generated audio prompts to reduce partial outputs and summarization.

Post-Transcription Translate Languages: Added French to the "Translate to" dropdown in audio mode.

UI Hint Refinement: Reduced helper-text size under the AI Model field so it fits cleanly in compact layouts.

Flash Model Upgraded: Replaced gemini-3-flash-preview with gemini-3.1-flash-lite-preview — Google's newer 3.1-generation model explicitly optimized for transcription and translation workloads, at lower cost ($0.25/$1.50 per 1M tokens vs $0.50/$3.00).

Removed Deprecated Model: Removed gemini-2.0-flash from the cost-tracking pricing table. Google deprecated this model and it will be shut down June 1, 2026.

Pricing Table Updated: Added accurate pricing for gemini-3.1-flash-lite-preview in the Google Apps Script cost calculator to ensure usage costs are tracked correctly.

API Token & Cost Tracking: Every Gemini API call now captures prompt tokens, completion tokens, and total tokens from the response metadata. These are logged to the Google Sheet alongside a new duration (seconds) column, enabling precise cost analysis per request. An automated Cost (USD) column is calculated server-side with a 10% safety margin, based on live Gemini pricing per model.

Error Message Logging: When a request fails, the specific error message is now recorded in the usage log, making it easier to diagnose issues without needing browser console access.

File Name Logging: Usage tracking now includes the original file name in every log entry. This makes it easier to identify which file was processed when reviewing the Google Sheet audit trail.

Usage Tracking: The system now logs every transcription, translation, and document translation event to a Google Sheet. Each log entry records the user, timestamp, mode, action, language, AI model, file type, file size, and success/failure status. This provides a full audit trail of system usage.

User Identification: The system now automatically detects the logged-in user from the Apache authentication header (X-Remote-User). No extra login step is needed — your .htaccess username is captured automatically. If auto-detection fails, a one-time email prompt appears as a fallback. The identified user is stored locally so it persists across sessions.

Process Another File: Added a "🔄 Process Another File" button in the results area. After completing a transcription or translation, you can now start processing a new file without reloading the page. Your language, recording type, model, and prompt selections are preserved.

Duplicate Translation Tabs: Fixed a bug where translating a second file would create orphaned tab elements from the previous transcription, causing duplicate tabs that couldn't be opened or downloaded. Translation tabs are now properly cleaned up between transcriptions.

Upload Progress Indicator: Added real-time progress bar with percentage display during file upload. Users can now see exactly how much of their file has been uploaded, especially helpful for large files.

File Preview & Validation: Instant file validation and preview after selection. The system now shows file name, size, format, and validation status (✅ ready, ⚠️ warning for large files, or ❌ error). Files are validated before processing starts.

Drag & Drop Upload: Enhanced file input with drag-and-drop zone. Users can now drag files directly into the upload area instead of clicking the browse button.

Real-time Form Validation: Form fields now show visual feedback (green checkmark for valid, red X for invalid) as you fill them out. The submit button is automatically enabled/disabled based on form completeness.

Session Recovery: Automatic session saving to browser storage. If you accidentally refresh the page or close the browser, your last transcription result will be restored within 24 hours.

Enhanced Error Messages: Error messages now include specific troubleshooting tips and suggested actions. Instead of generic errors, you'll see helpful guidance like "Check your internet connection" or "Try a smaller file" with specific file size information.

DOCX Translation UX: Document translations now display in the results area before downloading, allowing you to preview the translation. Previously, DOCX files would download without showing any preview.

Streamlined Translation Workflow: Removed the interrupting "Save Before Translating" modal. Translations now start immediately when you click the Translate button.

Better Upload Status Tracking: More detailed status messages during upload and processing stages, including "Uploading file...", "Upload complete!", and "Processing with AI..."

Format-Preserving DOCX Translation: Attempted to implement a new workflow to translate `.docx` files while preserving original formatting (headings, tables, etc.). The process involves parsing the document, translating text content, and reconstructing a new `.docx` file. Note: This feature is currently not working as expected and may produce corrupted or incorrectly formatted files.

AI Toolbox SPA Refactor: The application has been fundamentally refactored from a single-purpose tool into a multi-purpose "AI Toolbox" Single Page Application (SPA).

Document Translation Mode: Added a new "Document Translation" feature, accessible via a new tabbed navigation. This mode allows users to upload and translate text-based files.

Dynamic UI: The user interface now dynamically adapts based on the selected mode ("Audio Analysis" or "Document Translation"), showing only the relevant options and controls for each task.

Automatic Model Selection: The AI model is now automatically selected based on the "Type of Recording" to streamline the user workflow. 'KII' defaults to Gemini Flash, while 'FGD' and 'IDI' default to Gemini Pro. Users can still manually override the selection.

DOCX Download & Metadata: Downloaded .docx files are now dynamic, adapting to the language of the transcription and translation. They now include embedded metadata and dynamic Document Title (e.g., "Transcription-Nepali-YYYY-MM-DD").

Tab Functionality: Resolved a bug where the "Original" language tab was not clickable, preventing users from switching back to the original transcription.

Modal Button Logic: Fixed an issue where the "Go Back" button in the 'Save Before Translating' modal was incorrectly triggering a translation instead of closing the modal.

Dynamic Action Buttons: The "Copy" and "Download" buttons now dynamically update their text and function based on whether the user is viewing the original transcription or a translation, reducing ambiguity.

Results Section Revamp: The layout of the results area has been restructured for a more logical and intuitive workflow. The "Copy" and "Download" buttons now appear above the transcription text, and the translation controls are positioned directly below the text, improving the user journey from transcription to action.

Expanded Conditional Model Selection: The "AI Model" will now automatically switch to "Gemini Pro" when "FGD (Focus Group Discussion)" is selected, in addition to the existing rule for KII recordings.

Conditional Model Selection: The "AI Model" will now automatically switch to "Gemini Flash" when "KII (Key Informant Interview)" is selected as the recording type, streamlining the workflow for users.

Complete UI/UX Overhaul: Implemented a new, modern design system to improve user experience and align with Helen Keller Intl branding.

Enhanced Forms & Buttons: All form elements and buttons have been restyled with interactive hover effects, transitions, and clearer visual feedback.

Redesigned Header & Footer: The header and footer have been updated with gradient backgrounds and a more professional layout.

Styled Process Log: The process log now uses distinct colors and icons for different message types (error, success, info), making it easier to read.

Changelog Tracking: Added this changelog page to document all future updates to the system.