Furqan Shahid

I worked on vapi and make.com

Upwork • Freelance

Feb 2025 - Present • 1 yr 6 mos

Designed, built, and deployed AI voice agents for small and medium businesses using Vapi's voice orchestration platform. Responsible for end-to end agent development including STT integration with Deepgram, LLM orchestration with GPT-4o, and TTS integration with ElevenLabs. After building production voice agents on Vapi for over a year, I gained deep insight into the tradeoffs between flexibility and reliability in voice AI orchestration. Vapi's strength is in its flexibility you can bring your own STT (Deepgram, AssemblyAI), LLM (OpenAI, Anthropic, Groq), and TTS (ElevenLabs, Cartesia, PlayHT) providers, giving you complete control over the stack. However, this flexibility comes with complexity. Production costs range between $0.25-$0.33 per minute when factoring in all provider costs, with the $0.05/min platform fee being only a fraction of the total . Each provider requires separate billing, API keys, and fallback logic. I also experienced reliability challenges Vapi's status page documents multiple incidents including call worker degradation affecting 15,555+ calls, SIP gateway failures from disk space exhaustion, and calls going silent from deployment issues . The broader builder community reports consistent patterns of builders migrating from Vapi to Retell AI for production deployments where stability and ease of use become the priority . This experience led me to transition my production deployments to Retell AI, where the no-platform-fee model ($0.07/min base), consistent 620ms latency, and end-to-end pipeline reduce both cost and operational overhead