Shivam Kumar
A developer at VisionQuantech who created a zero-dependency open-source WhatsApp AI agent project.
Timeline (last 90 days)
After developer Shivam Kumar replaced the customer-service agent's model backend with DeepSeek in production, inference cost dropped by roughly 15x, while CSAT held steady across 4,000+ conversations.
Kumar stresses that model unit price is not the main source of cost surprises—doubling call volume is; he recommends logging token consumption at conversation granularity from day one.
Kumar opposes a wholesale model switch and recommends tiered routing: DeepSeek handles 85–90% of routine traffic (classification, drafting, retrieval QA, formatting), while frontier models handle the remaining 10–15% of complex tasks (ambiguous complaints, multi-step reasoning, low confidence).
An open-source project on GitHub provides a self-hosted WhatsApp AI agent, written using only the Python standard library with no pip package dependencies, supporting English, Hindi, and Hinglish, capable of lead qualification, FAQ answering, and appointment booking, and able to run locally in pure rule-based mode even without an API Key.
All Facts (4)
After developer Shivam Kumar replaced the customer-service agent's model backend with DeepSeek in production, inference cost dropped by roughly 15x, while CSAT held steady across 4,000+ conversations.
50%UnverifiedKumar stresses that model unit price is not the main source of cost surprises—doubling call volume is; he recommends logging token consumption at conversation granularity from day one.
50%UnverifiedKumar opposes a wholesale model switch and recommends tiered routing: DeepSeek handles 85–90% of routine traffic (classification, drafting, retrieval QA, formatting), while frontier models handle the remaining 10–15% of complex tasks (ambiguous complaints, multi-step reasoning, low confidence).
50%UnverifiedAn open-source project on GitHub provides a self-hosted WhatsApp AI agent, written using only the Python standard library with no pip package dependencies, supporting English, Hindi, and Hinglish, capable of lead qualification, FAQ answering, and appointment booking, and able to run locally in pure rule-based mode even without an API Key.
50%