MineMitra: Cosine Similarity Mining Q&A Bot
Sovereign Regulatory Authorities
How regulatory mining officers navigate state mining manuals and policies offline using a conversational bot matching natural language queries to vector indexes via cosine similarity.
Process Visualization
The Challenge
Mining inspectors in remote checkpoints and field offices needed instant access to dense regulatory manuals, tax structures, and compliance books. Due to lack of public internet and strict sovereign classification of regulatory datasets, commercial cloud-hosted large language model APIs could not be utilized.
The Solution
We built MineMitra, an offline semantic-search conversational agent. By parsing policy manuals into pre-structured QA pairs and converting queries into embeddings locally, the chatbot matches intents to database vectors using cosine similarity with zero cloud calls.
Implementation
Data formatting and indexing of state mining books into pre-structured QA datasets
Development of the local embedding pipeline using lightweight transformers
Vector index generation and cosine similarity lookup engine compilation
Desktop GUI packaging and cross-platform installation tests
Rollout to checkpoint workstations with offline database update scripts
Impact & Results
Operational Impact
- 95% accuracy in direct compliance query lookup matching
- Under 100ms response time per query fully offline
- Zero network dependency allowing continuous regulatory enforcement
Business Impact
- Reduced compliance query resolution time for checkpoint staff by 90%
- Standardized checkpoint enforcement interpretations across regional divisions
- Reduced administrative lookup errors and training cost
Strategic Impact
- 100% sovereign, self-contained regulatory knowledge base
- Data privacy shield securing government legal guidelines from external platforms
- Modular Q&A index framework supporting easy regulations updates