Tiger Agent for Bangladesh’s Leading Telecom Company

Industry: Telecommunications / Enterprise AI
Region: Bangladesh

Overview

Bangladesh’s largest telecommunications operator manages vast internal documentation — product policies, technical manuals, and operational procedures scattered across formats and silos with no unified access layer. Support and operations teams had no efficient way to retrieve precise answers, leading to slow resolution times and inconsistent responses.

Build NYC Tech designed and built Tiger Agent, a production-grade Retrieval-Augmented Generation (RAG) platform that transforms unstructured enterprise documents into an AI-powered knowledge assistant. It ingests PDFs and diverse document formats, converts them into vector embeddings stored in a self-hosted Weaviate database, and answers natural language queries using hybrid search.

The result delivers 92–96% retrieval precision with 6–10 second response times, full data sovereignty, zero per-seat licensing costs, and a security posture custom-built for regulated telecom environments.

Challenges

• Unstructured documentation: Internal knowledge was spread across inconsistent formats with no standardized structure, making automated extraction and reliable chunking difficult.
• Retrieval consistency: Pure semantic search underperformed on domain-specific telecom terminology, where exact keyword matches matter as much as conceptual relevance.
• LLM hallucination risk: Out-of-the-box LLM behavior required careful prompt engineering and temperature tuning to keep responses strictly grounded in source documentation.
• Query classification: The system needed to reliably distinguish between queries answerable from the knowledge base, from conversation context, or those genuinely unanswerable.
• Data sovereignty: All data had to remain within controlled infrastructure with no dependency on third-party hosted services and robust access control throughout.
• Enterprise operability: The client needed tools to manage documents, monitor usage, cache responses, and track performance without developer intervention.
  
  

Solutions

• Hybrid search with Weaviate: Combined vector-based semantic search with BM25 keyword scoring, tuned via configurable alpha weighting, to handle both conceptual and precise term queries across telecom content.
• Custom document processing pipeline: Built an automated ingestion pipeline (migrated from LangChain to a custom approach with PyPDF2) that parses, cleans, chunks, generates OpenAI embeddings, and loads documents into Weaviate with deduplication logic.
• Query rewriting: User queries are rephrased using conversation history before retrieval to resolve pronoun references and implicit context, improving accuracy across multi-turn conversations.
• Layered fallback architecture: A three-tier system routes queries first to the vector database, then to conversation history via embedding-based scoring, and finally to a defined “no result” response — preventing unsupported answers.
• LLM Judge prompt design: A self-evaluating prompt layer silently scores its own responses for quality using a structured feedback format, enabling continuous improvement without user-visible overhead.
• Self-hosted infrastructure: Weaviate, MongoDB, and Redis are fully containerized and available in both Docker and independent service configurations, giving the client complete data residency control and eliminating external licensing costs.
• Embeddable chat widget: A standalone React-based widget (orange-agent-chat) integrates into existing internal portals via iframe with no frontend rebuild required.
• Admin dashboard: A React + TypeScript admin panel covers document management, live conversation monitoring, query log analytics, suggestion management, response caching, and performance dashboards — fully operable by non-technical administrators.