Duvi research
Ideas for operational AI teams.
Practical writing on voice agents, customer automation, and the systems behind faster resolution.
Read the archiveManaging Lead Qualification Latency Across Distributed Channels
Learn how to eliminate response delays across WhatsApp, voice, and web channels by unifying business logic with AI agents.
Why Chat Prompts Fail on a Phone Call
Voice AI requires lower latency and shorter responses than text chat. Learn how to optimize system prompts and knowledge bases for phone calls using Duvi.
Why Customers Repeat Themselves Across Web and Voice Channels
Unified AI logic prevents context loss when customers move from web chat to phone support. Learn how a single knowledge base improves resolution rates.
Unifying Business Logic Across Voice and Text Channels
Coordinate your support logic across WhatsApp, web, and phone lines. Use a single knowledge base and API connectors to ensure consistent customer experiences.
Moving Beyond Static Answers to Actions Driven by APIs
Learn how to move AI agents beyond static documentation by connecting them to APIs and Salesforce for real time customer resolution on web and phone channels.
The Gap Between Answering a Question and Solving a Ticket
Moving AI agents beyond search and into execution. Learn how Duvi connects to APIs and Salesforce to solve tickets instead of just answering questions.
Knowledge Drift and Why Phone Lines Struggle to Match Website Content
Knowledge drift occurs when voice and web channels use separate logic. Learn how unified AI agents maintain consistency across every communication channel.
Replacing the Phone Tree with Contextual Intent
Replace rigid phone menus with AI voice agents that understand natural language and execute tasks through APIs.
Connecting Support Agents to Live Business Data
Connecting AI agents to Salesforce and APIs allows for real time resolution of customer inquiries using live business data instead of static FAQs.
The Cost of the Internal Ticket Transfer
Internal ticket transfers cost roughly twenty dollars per incident in labor. Reducing the friction between knowing and doing requires AI agents capable of API tool calls and record updates.
Building the Mahindra Electric Agent: Architecture
How we built the customer agent for Mahindra Electric: four services split by data ownership, why enquiry and booked are separate, and what handover must carry.
Prompt Caching Across Turns, and Its Sharp Edges
How we key a prompt cache, why we extend at half life, why creation sits behind a claim lock, and the rule: a cache may save money, never cost time.