Agentflo

Agentflo is an AWS-powered agentic AI platform that enables FMCGs to deploy voice-enabled CPaaS-integrated sales agents that automate order booking, campaigns, and retailer engagement at scale.

Comment

Inspiration

Ten years ago, we helped modernize Pakistan’s FMCG industry by building Salesflo, a platform that digitized field sales, order booking, and route-to-market visibility across the country. But a decade later, we saw many inefficiencies resurface: manual order flows, sales teams stretched thin, retailer engagement dependent on availability, and decision-making disconnected from data.

At the same time, the world was entering a new era: agentic AI. While global industries were shifting toward autonomous AI-driven workflows, Pakistan and similar markets were still operating in pre-AI cycles. We realized that this wasn’t just a technological gap, it was a chance to redefine the future of FMCG engagement once again.

This drove us to create Agentflo: a platform that gives every retailer their own AI-powered sales agent that operates 24/7, speaks their language, remembers their behavior, and drives real commercial growth.

What it does

Agentflo is a B2B agentic AI platform that allows FMCG companies to deploy fully autonomous, voice-enabled AI sales agents that operate across multiple CPaaS channels like WhatsApp, Wechat and voice calls.

These AI agents can:

  • Place and manage retailer orders via natural conversation.
  • Recommend SKUs using purchasing trend analysis.
  • Run targeted campaigns based on retailer segments (e.g., churned, new, reactivated).
  • Remember preferences such as pack size or frequently ordered items.
  • Use semantic product search to match Natural Language Queries with exact SKUs.
  • Automatically send catalogs, promotions, or invoices to the retailer via WhatsApp templates. -Smartly upsell, cross sell, meet sales targets and much more.

The platform includes a web portal for full control and observability over your customer segments, campaigns and a no-code agent builder, enabling companies to configure personalities, behavior flows, target segments, and launch campaigns. to put it shortly its almost like a "Shopify for AI sales agents."

How we built it

As part of the AWS Partner Network, we built Agentflo using AWS-native services to ensure reliability, familiarity, and scalability from day one. We used AWS AgentCore as a backbone for agent orchestration and concurrency management, and utilized services like Strands to support large-scale execution.

For this hackathon, we built:

  • A WhatsApp-based agentic interface powered by real-time voice capabilities.
  • A low-latency inference pipeline for voice interaction.
  • A no-code web interface for businesses to deploy their own sales agents.
  • A semantic vector search system for natural product discovery.
  • SKU recommendation models based on retailer ordering patterns.
  • A scalable infrastructure capable of handling heavy concurrency.

This stack and the agentcore runtime deployment allowed us to handle upto: -500 concurrent users -48,000 users per day -100 invocations per second -100mb payloads -8 hour long sessions

Challenges we ran into

Challenge How we solved it
Voice response latency was initially 10–20 minutes Optimized the pipeline down to a few seconds
Retailers used vague, natural-language product names Implemented vector-based semantic product search
Creating production-level scale within hackathon time Leveraged AWS AgentCore for concurrency and reliability
Ensuring agents felt role-specific and human-like Introduced configurable personalities and interaction flows
Multilingual user expectations Adapted dynamic language intent handling

Accomplishments that we're proud of

  • Built a fully functional, voice-enabled AI sales agent prototype within hackathon time.
  • Launched a no-code AI agent builder for FMCG companies.
  • Achieved enterprise-level concurrency and latency benchmarks.
  • Delivered a seamless WhatsApp-based product ordering flow using semantic intelligence.
  • Secured production pilots with Nestlé Pakistan and EBM.
  • Attracted interest from FMCG companies in Saudi Arabia, Malaysia, and Australia.

What we learned

  • Agentic AI requires a combination of memory, orchestration, and goal-driven behavior.
  • Latency is critical in voice-based experiences; even minor delays impact trust.
  • Retailers prefer conversational engagement over dashboards or apps.
  • AWS-native infrastructure significantly accelerates production readiness.
  • Local-language fluency and contextual reasoning are key for adoption in emerging markets.

What's next for Agentflo

We believe the future of FMCG sales is powered by a hybrid model:

$$ \text{Human Strategy} + \text{Agentic AI Execution} = \text{Scalable Commercial Growth} $$

Our next steps include:

  • Scaling globally across multilingual FMCG markets.
  • Expanding agent personalities and domain expertise.
  • Enhancing campaign intelligence with predictive behavior modeling.
  • Deeper integrations with CRMs and e-commerce systems.
  • Launching a marketplace for pre-built AI sales agent templates.

Agentflo is our next step in putting Pakistan at the forefront of agentic AI-driven commerce — and this time, we’re building it for the world.

Built With

Share this project:

Updates