Techtimize
TECHTIMIZE

AI-Native Engineering

Initializing AI stack…

GovernmentAI & LLM Integration

2M monthly citizen queries, 78% query deflection from call centre

IndustryGovernment / Public Sector
Team11 engineers
Timeline20 weeks

The Challenge

A GCC government ministry received 2.6M citizen enquiries monthly through a call centre — 78% of which were routine questions answerable from public policy documents, regulations, and procedural guides. Call centre costs exceeded SAR 18M annually. An existing chatbot had a 34% accuracy rate and was widely distrusted by citizens who preferred to call instead.

The Solution

TECHTIMIZE built a production RAG system on AWS Bahrain (me-south-1) ingesting 4,200 policy documents, regulations, and procedural guides across Arabic and English. Pinecone vector store with hybrid retrieval (dense + sparse) achieves 99.2% Recall@5 on evaluated query sets. GPT-4 generates cited, contextual responses grounded in official documents. The system is deployed as a web widget embedded on the ministry portal and WhatsApp Business API — all data remaining within Saudi Arabia under NCA ECC controls.

Technology Stack

Node.jsReactPineconeOpenAI GPT-4LangChainAWS BahrainArabic NLP

Results

0
Monthly Queries
Citizen queries handled by AI knowledge base per month
0
Query Deflection
Call centre enquiries deflected to AI — 2M calls/month reduced to 440K
0
Response Accuracy
Answer faithfulness on evaluated query set (RAGAS metric)
0
Annual Savings
Projected annual call centre cost reduction

How We Delivered It

1
Document Corpus Preparation
Ingested, cleaned, and structured 4,200 Arabic and English policy documents with semantic chunking and metadata tagging.
2
Multilingual Vector Store
Built Pinecone index with multilingual-e5-large embeddings, supporting Arabic, English, and mixed-language queries.
3
Hybrid RAG Pipeline
Implemented dense + BM25 sparse hybrid retrieval with cross-encoder re-ranking, achieving 99.2% Recall@5.
4
Arabic Response Generation
Tuned GPT-4 prompts for formal Arabic government communication style with citation requirements for every response.
5
NCA ECC Compliance
Deployed entirely on AWS Bahrain (me-south-1). All citizen data stays in-country with full NCA ECC audit logging.
6
WhatsApp & Web Integration
Deployed as ministry website widget and WhatsApp Business API integration. 54% of queries come via WhatsApp.
"Citizens now get accurate, cited answers to their questions in seconds — in both Arabic and English. The trust in our digital services has fundamentally shifted."
Director of Digital Transformation
GCC Government Ministry
Back to Case Studies

Ready for Similar Results?

Let's talk about your challenge. Our team delivers measurable outcomes fast.

Start a Project →More Case Studies