Dev Seth
← All work
GovernmentIn production

Round-the-clock citizen service, with intelligence on every conversation

A bilingual voice AI assistant for a government entity in the Middle East that answers citizens 24/7 and shows leadership what people are actually calling about.

Citizen callSpeech recognition(AR/EN)Voice agentResponseConversationintelligenceManagementassistant
24/7

service, up from 12 hours a day

200+

enquiries handled every day

Arabic & English

in both languages

~$52K

saved every month

the situation

Citizen enquiries were handled manually by a 10-person team, and only during a 12-hour window. Leadership had little visibility into what people were calling about.

the hard part

It had to handle real conversations in two languages at government-grade reliability, and do more than answer calls: management needed to understand the conversations, not just clear the queue.

what i built

01

Bilingual voice agent

Handles enquiries in Arabic and English around the clock.

02

Conversation intelligence

Tracks what is being discussed across calls, so it's clear who is talking about what.

03

Management assistant

Surfaces those insights to leadership, so decisions are based on what citizens actually raise.

how it fits together

Citizen callSpeech recognition(AR/EN)Voice agentResponseConversationintelligenceManagementassistant

results

  • Service moved from 12 hours a day to 24/7.
  • 200+ enquiries handled daily, in Arabic and English.
  • Took over the workload of a 10-person team, saving about $52K a month.

stack

Speech-to-text & text-to-speechLLM agentsConversation analytics

next case study

RetailIn production

Predicting stock-outs before they happen

An AI agent for a UK retailer with 22+ stores and 5 warehouses that predicts stock-outs from events and demand signals, and recommends what to move where.

$65Ksaved every month

22+ storesand 5 warehouses

Stores (22+)Warehouses (5)Central datapipelineEvent-awareforecastingStock-outpredictionRecommendations:what to move where
Read the next case study →

Open to partnerships and select projects.

If you're building something where AI has to work in production, or has to stay private, I'd like to hear about it.