Skip to content
<coderuzb />
21 Apr 2026

ai-xodim.uz — AI Agents Catalog for Businesses in Uzbekistan

ai-xodim.uz is a platform catalog of AI employees for businesses in Uzbekistan: AI salespeople, operators, and assistants, integrated with OpenAI LLM.

ai-xodim.uz — AI Agents Catalog for Businesses in Uzbekistan

Context and Goal

AI technologies are growing rapidly in Uzbekistan, but businesses struggle to understand which solution actually fits them. Agents are scattered across different vendors, pricing is opaque, and comparing them is hard.

The goal of ai-xodim.uz was to build a single catalog storefront of AI employees for Uzbek companies: a client picks an AI salesperson, call-center operator, HR assistant, or support agent in one place, requests a demo, and connects it to their business.

Key requirements:

  • Interface in 4 languages (uz, oz, ru, en)
  • Agent categorization and filtering
  • Live demo chat powered by OpenAI
  • Convenient admin panel on Filament

Tech Stack and Architecture

The project is built as a monolithic Laravel application — a good fit for fast iteration and a small team.

  • Backend: Laravel 11, PHP 8.3, Queues (Redis), Horizon
  • Admin panel: Filament 3 — resources, relation managers, custom pages
  • Frontend: Blade + Alpine.js + Tailwind CSS — a lightweight SPA-like experience
  • AI layer: OpenAI Chat Completions API via openai-php/client
  • Database: MySQL with spatie/laravel-translatable for 4-language content
  • Cache and queues: Redis

Key architectural decision: all LLM-related logic lives in the Services\\AI namespace. Each agent type has its own prompt template, so adding a new agent means one DB record and one prompt file — no code changes.

$reply = app(AgentChatService::class)
    ->for($agent)
    ->ask($userMessage);

Key Features

  • AI agent catalog — organized by category (sales, operator, HR, support, marketing)
  • Live demo chat — users chat with an agent directly on the site via OpenAI
  • Filters and search — by industry, price range, integrations (Telegram, WhatsApp, CRM)
  • Multilingual content — 4 languages, edited in a single admin form
  • Lead form with CRM webhook — demo requests are forwarded automatically
  • Filament admin panel — resources for agents, categories, leads, plans, and prompts
  • Analytics — views, demo sessions, and conversion per agent
  • SEO — server-side rendering, 4-language hreflang, sitemap.xml, structured data

Challenges and Solutions

1. OpenAI integration and cost control. A live demo chat means constant API calls — real money. Solutions:

  • Rate limiting per IP and session (Laravel RateLimiter)
  • Per-agent max_tokens and model settings — cheap gpt-4o-mini as default
  • Redis caching of prompts and responses for repeated questions
  • Async handling of long responses via queues

2. Content and prompts in 4 languages. Each agent had to speak in its own voice in every language. Using spatie/laravel-translatable, I made prompts translatable too: the admin fills in four versions in a single form.

3. Tight deadline (1 month). A Filament-first approach: build the admin panel and data model first, then the public site. The client could start filling the catalog before the frontend was finished.

4. Preserving chat context. In the demo chat, users ask several questions in a row. For each agent session, the last N messages are stored in the DB and sent as context with the next request, while older messages are trimmed automatically.

Results

  • MVP launched in 1 month, live on the public domain ai-xodim.uz
  • Content in 4 languages — each agent card and prompt in 4 versions
  • Average AI response time: 1.5–2.5 seconds (with streaming)
  • Average page load: under 1 second (cache + simple Blade + Alpine)
  • Filament admin handed off to the client's marketing team — new agents are added without a developer
  • Demo leads flow straight to CRM with no manual re-entry

The key outcome for the client: a first single-window platform for AI solutions on the Uzbek market.

Takeaways and Roadmap

ai-xodim.uz is a catalog with room to grow into a full AI-agent marketplace. Next steps:

  • User accounts — companies rent an agent and tune it for themselves
  • Personalized prompt builder based on the client company's data
  • Ready-made Telegram and WhatsApp Business integrations as templates
  • RAG layer — plugging the client's knowledge base (docs, FAQ) into the agent
  • Payments and subscriptions — Click, Payme, Uzcard
  • Voice agents with STT/TTS

On the technical side, the stack proved to be a good fit: the Laravel + Filament + Alpine combo lets a small team move fast and scale the product step by step.

← All work
Related work