https://fathom.video/share/BtiYTVNTtMkyk42drE2BrsZSjtQzBxqF

VIEW RECORDING - 36 mins (No highlights)

. Meeting Purpose

Technical discovery for integrating Valsea's AI voice solution into Coca-Cola's systems. This conversation took place to explore the possibility of integrating Valsea's AI voice-based solution with Coca-Cola's systems, specifically for the Zalo MiniApp and cooler reporting, and to understand the technical requirements and potential workflows.

3. Collaboration Opportunity

Potential ways the organization could collaborate with VALSEA:

• Primary Use Case (Zalo MiniApp Integration): Voice-enable the new Zalo MiniApp to automate ordering for Coca-Cola's ~400k active retailers.
  • Workflow: Retailer voice → Zalo MiniApp → Valsea ASR → Structured JSON → SAP ERP.
  • Coca-Cola has ~300 SKUs; an A/B test is suggested to determine if fine-tuning Valsea's models for product names is needed.
  • Valsea's AI solution offers transcription, translation, sentiment analysis, and data annotation to extract structured data (SKUs, addresses) from voice.
• Secondary POC (Cooler Reporting Automation): Automate reporting for Coca-Cola's 150k–200k coolers via a voice channel.
  • Goal: Replace manual reporting currently done via a BPO call center or Zalo fan page with a voice-based system for retailers to report issues (e.g., cooling, placement, customer complaints).
  • Future Possibility: Explore using Agora's RTT services for SIP landline integration, leveraging Valsea's existing POC with Agora.
• Technical Integration: Valsea's API is OpenAI SDK-compatible and supports both pull and push methods.
  • Preferred Method: A push-webhook architecture where FPT sends audio to Valsea, keeping FPT's systems closed to external access.
  • Latency: Approximately 200ms RTT.
  • Hosting: Flexible options (AWS Singapore, GCP, Alibaba Cloud, on-premise in Vietnam) to minimize latency.

4. Speech / Language Challenges Discussed

• Core Technology: Valsea's ASR & TTS solution offers a 10–30% lower Word Error Rate (WER) for Vietnamese compared to competitors (OpenAI, 11 Labs, Google).
• Why Valsea is Better: This improvement is achieved via on-premise training on local Vietnamese dialects, slangs, and accounting for background noise common in market environments.
• Specific Challenges Addressed:
  • Multilingual/Accented Speech: Understanding diverse Vietnamese accents (northern, central, southern).
  • Slang: Translating incoherent slang into understandable business language (e.g., in customer calls).
  • Background Noise: Performing accurately in noisy environments like markets.
  • Workflow Automation: Converting voice messages (e.g., from busy shop owners) into actionable JSON data for automated system input, applicable in logistics (fleet management, warehouse management) and sales.

5. Strategic Fit

• Valsea's specialized ASR for Vietnamese, with its high accuracy, directly addresses Coca-Cola's need for efficient voice-enabled ordering and customer service in the Vietnam market.
• The collaboration can help Coca-Cola automate repetitive tasks, improve order processing efficiency, and enhance customer experience, as highlighted by Rahul's interest in automating processes for ~150k-200k coolers.
• FPT, as the system integrator, sees a clear path for technical integration and can expand its service offerings by leveraging Valsea's specialized AI capabilities.
• The project could serve as a case study for both Valsea and FPT for similar solutions in the FMCG sector and other industries requiring localized voice AI.