AIX | Devpost

order tab
inventory tab
omnichannel tab

Inspiration

Across the MENA region, millions of small businesses sell primarily through social media DMs on WhatsApp, Instagram, and Facebook. These businesses don’t fail because of bad products they fail because they miss messages, reply too late, or can’t scale human sales staff 24/7.
We repeatedly saw merchants lose real revenue simply because no one replied in time, stock information was outdated, or orders were handled manually and inconsistently.

With the launch of Gemini 3 and its focus on the Action Era, we saw an opportunity to move beyond chatbots and build something closer to a digital employee an AI that doesn’t just talk, but actually operates. That idea became AIX.

What it does

AIX is a fully autonomous AI Sales Agents that automates social commerce end-to-end.

It:

Replies instantly 24/7 across WhatsApp, Instagram, Messenger, and Telegram
Understands Arabic, English, and Franco-Arab naturally
Extracts structured orders directly from free-form customer messages
Checks and syncs inventory in real time to prevent overselling
Handles clarifications (size, color, quantity), follow-ups, and confirmations
Centralizes all chats, orders, and inventory in a unified dashboard

In practice, one AIX agent can replace 2–3 human sales staff, eliminating missed messages and abandoned DMs.

How we built it

AIX is built as an autonomous, tool-using agents powered by the Gemini 3 API.

At its core:

Gemini 3 handles multilingual reasoning, intent detection, and multi-turn conversation planning
Structured tool calls are used for inventory checks, order creation, and status updates
The agents maintain conversational and operational state across long-running sessions
A multichannel ingestion layer unifies messages from Meta platforms and Telegram
A SaaS dashboard exposes chats, orders, inventory, and human takeover controls

Rather than a prompt-only wrapper, AIX is an orchestrator where Gemini 3 plans actions, validates outcomes, and adapts dynamically based on real business data.

Challenges we ran into

Unstructured language to structured orders: Customers describe orders inconsistently, mix languages, and change their minds mid-chat.
Franco-Arab parsing: Handling Arabic written in English characters required careful intent normalization.
State management: Conversations span hours or days and must stay consistent with inventory changes.
Trust and safety: Ensuring the AI confirms ambiguous orders instead of making assumptions.
Latency expectations: Social commerce requires near-instant replies without sacrificing reasoning quality.

Accomplishments that we're proud of

Built a truly multi sales agents, not a chatbot
Achieved reliable order extraction from messy, real-world DM conversations
won 2nd place across MENA region in Huawei Developer Competition
Enabled real-time inventory synchronization across multiple channels
Designed a system where AI-first automation and human takeover conversation smoothly
Created a product that delivers immediate, measurable business value

What we learned

Autonomy matters more than fluency businesses care about outcomes
Validation over perfection the faster we deliver and test the better the product will be not just testing
Long-context, multi-step reasoning is essential for real commerce workflows
Multilingual AI in emerging markets must handle informal, mixed-language inputs
AI agents need guardrails, validation loops, and state not just prompts

Most importantly, we learned that the future of AI in business is operational, not conversational.

What's next for AIX

Next, we plan to:

Expand autonomous capabilities (returns, refunds, upsells, post-sale support, analytics)
Add self-optimizing sales strategies driven by performance feedback
Support more regional dialects and vertical-specific flows
Integrate deeper with logistics and payment providers
Scale AIX as the default AI sales layer for SMBs across MENA

AIX is not just automating replies it’s building the foundation for AI-run commerce operations.