This release focuses on monitoring transparency, high-speed data enrichment, and significantly improved competitor detection accuracy in multilingual environments.

1. Monitoring UX Improvements & Model Expansion

We have addressed critical scheduler issues and expanded the available model suite to provide a more reliable monitoring experience.

Monitoring Cycle Fix: Resolved an issue where ChatGPT tests failed to complete full cycles, leading to incorrect "Next Run" delays of up to 324 hours. The system now guarantees full test completion per cycle.
Countdown Timer: A real-time "Next Update" countdown has been added to monitoring pages, providing users with clear visibility into data refresh schedules.
DeepSeek Integration: DeepSeek is now available as a model option in the monitoring UI, allowing users to track brand presence in this rapidly growing AI ecosystem alongside GPT and Gemini.
Data Visualization Polish: Fixed misleading partial data in competitor mention charts during active monitoring cycles.

2. Enrichment Pipeline Overhaul

We have eliminated the "social scoring bottleneck" that previously caused significant delays when adding prompts in bulk.

Parallel Processing with BullMQ: The enrichment pipeline has moved to a worker-based architecture. By decoupling social scoring from the main request and using parallel processing, we have drastically reduced wait times for Reddit API analysis.
Live Status Insights: Using SSE (Server-Sent Events), the UI now provides real-time updates for each row. Users see live status indicators (e.g., "Extracting keywords...", "Analyzing Reddit...") instead of pending states.
Asynchronous Execution: Social scoring no longer blocks prompt creation. The process runs entirely in the background, supported by real-time enrichment banners and toast notifications.

3. Advanced Analysis Engine Refinement

We have implemented a hybrid analysis approach to solve data gaps in competitor extraction and source classification, particularly for DeepSeek and Arabic-language content.

Enhanced Competitor Detection: To address a 70% miss rate in LLM-only extraction, we implemented a hybrid "LLM + Regex" fallback. This ensures English brand names embedded in Arabic text and niche competitors are correctly identified.
Precise Source Classification: Improved the classification logic for software directories (e.g., SoftwareSuggest), mobile app stores, and regional Arabic product pages. This reduces error rates in source attribution by approximately 45%.
Provider-Specific Strategies: Tailored classification strategies have been deployed for DeepSeek, GPT, and Gemini results to account for the unique way each model structures its citations.