Choco

Solution

Choco rebuilt its food-distribution platform around OpenAI APIs so order capture could move from workflow support to execution. OrderAgent ingests multimodal inputs including emails, SMS messages, images, and documents, then uses transcription, extraction, structured outputs, function calling, and customer-specific context from prior ordering behavior to convert them into ERP-ready orders. VoiceAgent extends the same model to phone ordering through the Realtime API with sub-second latency, enabling distributors to accept spoken orders after hours while maintaining continuity with the rest of the ordering stack. Choco also built evaluation infrastructure with ground-truth datasets, continuous monitoring, and A/B testing so the system can decide when to trigger an optional Autopilot mode and when to keep humans in the loop. The result is a context-aware operational agent that can resolve ambiguity at order capture instead of relying on staff to translate every incoming request by hand.

Data flow

Orders arrive as emails, texts, images, documents, or phone calls and are routed through OrderAgent or VoiceAgent. OpenAI models transcribe, extract, and resolve ambiguous items against customer-specific history and catalogs, then produce ERP-ready orders; Autopilot can finalize confident cases while low-confidence cases go to human review.