Every promoted item is checked against the source of truth, with problems surfacing in minutes, not weeks.
Wegmans Marketplace Validation Agents
AI Exchange Hackathon
SynapseAI B2B Wholesale
Wholesale buyers place and change orders in plain language, and see the price they will actually pay.
Conversational AI for B2B wholesale ordering, demand forecasting, and campaign generation.
What the agent does
Takes a wholesale order in conversation, forecasts demand, drafts the promotion, and holds each one for human approval.
- Brand
- Mars
- Contributed by
- MACH Alliance (Hub operator)
- Industry
- CPG, Retail
- Audience
- B2B
01
The problem
Why this exists
Wholesale buyers at consumer-packaged-goods brands place complex, repetitive orders against constantly shifting inventory and promotional rules. The buyer wants to ask "give me my usual but adjust for next month's launch" and have the system understand. The seller wants the buyer to discover new SKUs, qualify for the right promotion, and submit a clean order without a human re-entering it.
Read more
The current path is portal forms, EDI files, account managers on the phone, and email. Every layer adds latency and error. Promotion rules change weekly and only the sales team knows them. New SKUs get missed. Demand signal lives in a spreadsheet that nobody syncs back to the planning team.
02
Outcomes
What this demonstrates
- A wholesale buyer can place, modify, or query an order in plain language across web and WhatsApp.
- Promotional eligibility is computed in the loop, so the buyer sees the price they will actually pay before they confirm.
- Demand signal from buyer conversations feeds back into AWS Forecast, sharpening the brand's own planning rather than living in someone's inbox.
Read more
- Campaign drafts are generated against real account context, so a marketing manager can ship a personalized campaign to a buyer in minutes rather than days.
- High-value orders and campaign drafts pause for a human approval before anything runs, with the audit trail kept for compliance.
- The whole stack runs on a composable architecture (commercetools, Voucherify, custom agent layer) so brands can adopt SynapseAI without giving up the platforms they already invested in.
03
How it works
Architecture
SynapseAI is built around LangGraph, which orchestrates a stateful conversation across the components a B2B buying agent needs. The graph routes user intent to the right tool: commercetools handles products, carts, and orders; Voucherify handles loyalty and promotion logic; AWS Forecast handles demand prediction; the LLM, accessed via LiteLLM against AWS Bedrock so model swaps are config changes, handles reasoning and text generation.
Read more
The backend is a FastAPI service with auto-generated Swagger docs. Session state and short-lived agent memory live in Redis. The agent uses the Model Context Protocol (MCP) to expose its capabilities to outside callers, including a WhatsApp integration that lets a buyer place an order from their phone without ever opening a portal.
The frontend is a React 19 application built with Vite and TanStack Router. The chat UI is the primary surface but the same backend serves a more traditional dashboard for sales managers who want to see what the agent is doing on their behalf.
Infrastructure is fully described in Terraform: AWS ECS Fargate for the API, an Application Load Balancer in front, AWS ElastiCache for Redis, and AWS ECR for container images. The documentation site's source sits in the public repository, alongside Terraform deployment guides for brands that want to stand up their own instance.
Tech stack
LangGraphAWS BedrockLiteLLMFastAPIPythonRedisMCP (Model Context Protocol)React 19ViteTanStack RoutercommercetoolsVoucherifyAWS ForecastAWS ECS FargateTerraform
Next step
Take the code
This one is published in full. Read it, run it, fork it. Nothing here is behind a conversation.
Request an intro
This one is the MACH Alliance’s own work. Ask about conversational wholesale ordering, and the people who built it will answer.
Not ready to talk to anyone? MACH Insider is the Alliance mailing list. New solutions, program openings, and the research behind them, once a month.
Related solutions
Copyright claims are classified against past cases in seconds instead of days, with an analyst deciding.
Fraud Detection & DMCA Takedown AI
AI Exchange HackathonIn six weeks, Dr. Martens and Conscia built a prototype on-site shopping assistant for drmartens.com that could search, size, and add to cart.
Dr. Martens On-Site Shopping Assistant
AI Exchange Hackathon
Verification
Public code
The implementation is published in full, so anything on this page can be checked against the repository. Anything the repository does not show is not established here.
Status: Reference prototypeAdded May 26, 2026A listing is an editorial decision, not a MACH Alliance certification.
Built something worth showing?
Put your agentic solution in front of the enterprise leaders who use the hub to decide what to build next. Submissions are open to MACH Alliance Members, Collaborators, or Ambassadors.