Solving B2B Deal Amnesia with AI: Building cheersthepure DispatchAI

Solving B2B Deal Amnesia with AI: Building the cheersthepure DispatchAI

In the fast-paced world of FMCG (Fast-Moving Consumer Goods) and beverage distribution, verbal agreements are the lifeblood of wholesale business. At cheersthepure, a packaged drinking water manufacturer, managing orders across hundreds of local distributors, supermarkets, and event caterers exposed a massive operational bottleneck: deal amnesia.

When a distributor negotiates a custom volume discount (e.g., 5% off on 1L crates) or specifies a strict delivery constraint (e.g., "our warehouse is inaccessible before 3 PM"), this context is often lost across disconnected chat logs and manual CRMs. The result? Disputed invoices, delayed dispatches, and friction with wholesale buyers.

For the HackwithHyderabad 3.0 hackathon, our team set out to solve this exact problem. Instead of building a standard FAQ chatbot, we engineered the cheersthepure DispatchAI—an autonomous B2B supply and deal intelligence agent that actually learns from its buyers.

The Tech Stack: Giving AI Persistent Memory

To move beyond stateless conversations, we required an architecture capable of cognitive memory. We built our solution using:

• Vectorize Hindsight: A persistent memory system that allows the agent to retain, recall, and improve over time rather than forgetting past sessionsspan_0span_0span_1span_1.
• Groq: Ultra-fast LLM inference utilizing qwen/qwen3-32b for rapid reasoning and responsesspan_2span_2span_3span_3.
• Django & Python: Our core backend routing layer that connects the B2B portal to the AI agent.

How It Works in Practice

The true power of the DispatchAI lies in its Hindsight-powered learning curvespan_4span_4span_5span_5.

1. The Cold Interaction (Retain)
A distributor messages the system: "We need 120 cases of 1L bottles. Please note our access road is undergoing construction, so deliver only after 4:00 PM."
The agent confirms the order and triggers a retain operation, committing this logistical constraint to the cheersthepure-b2b memory bank.

2. The Warm Return (Recall)
Two weeks later, the same distributor returns: "Need a restock of 150 cases."
Before generating a response, our backend triggers a recall operation to pull context from days or weeks agospan_6span_6span_7span_7. The agent seamlessly replies: "Welcome back! Rebooking 150 cases of 1L. Are we still scheduling delivery strictly after 4:00 PM due to your access road construction?"

3. The Business Moat (Dispute Resolution)
If a buyer attempts to claim an unverified verbal discount ("Apply the 10% discount you promised last time"), the agent cross-references its memory bank. It politely upholds standard pricing if no such agreement exists in its retained history.

The Real-World Impact

By treating AI as an operational logistics partner rather than a simple support widget, cheersthepure DispatchAI eliminates revenue leaks caused by forgotten deals. Equipping an agent with persistent,cross-session memory turns it from a novelty tool into a core business asset that companies would pay to adoptspan_8span_8span_9span_9.