A leading Middle Eastern mining enterprise operating multiple extraction and processing sites. With a strong focus on digital transformation, the client sought a conversational interface to simplify access to operational KPIs for executives and stakeholders

• Technical Barrier to KPIs: Operational managers and executives could not access real-time or historical data without relying on technical teams for manual database queries.
• Data Fragmentation and Context: Data was scattered across disparate mining systems, making efficient retrieval of cross-site or trend data difficult, especially when requiring context from internal reports.
• Query Reliability: Initial attempts at Text-to-SQL struggled with complex mining schemas, leading to inconsistent and unreliable Large Language Model (LLM) generated outputs.
• Cross-Platform Accessibility: The solution needed to support natural language queries via both voice and text across mobile platforms for users in remote locations.
   

QBurst developed a mobile-first GenAI chatbot solution with voice and text interfaces, seamlessly integrated with the client’s existing backend systems. The solution focuses on providing real-time, natural language access to operational KPIs through a sophisticated Hybrid Query Architecture.

Key elements of the architectural approach:

• Data Integration: Established REST APIs over existing KPI databases and built a Python-based transformation layer to unify and cache metrics across all sites.
• Knowledge Base (RAG): Processed internal mining reports using tools like LangChain and stored vector embeddings in ChromaDB. This Retrieval-Augmented Generation (RAG) system grounds the AI's responses in internal documentation.
• Hybrid Query Engine: To ensure reliability, the system combines controlled Text-to-SQL translation with the RAG context engine. This uses mining documentation and schema metadata to enhance prompt accuracy, ensuring high precision even for complex structured data requests.
• AI Model Strategy: Leveraged GPT-4 Turbo for high-accuracy, complex use cases and GPT-3.5 Turbo for cost-efficient queries, applying advanced prompt engineering techniques to simulate expert reasoning.
   

• Hybrid Query Engine: Combines the flexibility of LLMs with the reliability of pre-tested SQL templates, achieving high precision through schema-aware, context-enriched prompts.
• Vector Search Implementation: Utilized vector search in ChromaDB to quickly identify relevant document chunks, which are then injected into the OpenAI prompt for contextual grounding.
• Mobile-First UI: Built with Flutter for cross-platform mobile support, integrating voice-to-text, multilingual capability, and visual KPIs.
• Security and Governance: Enforced fine-grained Role-Based Access Control (RBAC) across the platform and secured all API endpoints with robust authentication.
   

The GenAI solution delivered transformative improvements in executive decision-making and operational efficiency:

• Accelerated Executive Decision-Making: Instant voice and text access to critical KPIs accelerated strategic planning, saving stakeholders an estimated 55% of weekly time previously spent navigating database systems.
• Reduced Operational Costs: Automated reporting and data processing significantly reduced manual data effort, saving an estimated 45% of monthly person-hours and minimizing training requirements for new users.
• Strengthened Compliance: By providing instant, standardized access to operational metrics, the solution reduced the time required for government audit data preparation by 60%.
• Smarter Strategic Planning: Comprehensive access to historical trends and real-time performance metrics across all mining sites supports data-driven growth decisions.
• Future Innovation: Established the infrastructure for advanced analytics, predictive insights, and AI-driven operational optimization across mining operations.