/ Work · Voice AI
An enterprise voice AI platform built for Natsoft, a US consulting and BPO enterprise with 2,000+ employees. It started with recruitment calls and grew to insurance, sales and support.

The client
Natsoft
Recruiters were buried in repetitive follow-ups, candidates dropped off after offers, and nobody captured why people rejected them.
The problem
Replace repetitive, high-volume human calling with AI conversations that still feel human, and send structured results back to the CRM.
We proposed a configurable voice platform rather than a single-purpose bot, so HR and operations teams could adapt it without engineering.
What we built
Visual flow builder
- Drag-and-drop nodes, prompts, logic paths and fallbacks, built for non-engineers
Swappable voice stack per flow
- Speech-to-text: Deepgram, Google STT, AssemblyAI, Gladia
- Language models: OpenAI and Google
- Text-to-speech: ElevenLabs and Google TTS
- RAG against SOPs, FAQs and policy documents
CRM loop and analytics
- An offer released in the CRM triggers the outbound call
- Recordings, transcripts and sentiment sync back
- Per-call LLM and TTS latency, hang-up reasons and live transcripts
Stack Python · FastAPI · OpenAI · Deepgram · ElevenLabs · WebSockets · React · PostgreSQL



Outcome
A recruitment tool that became an enterprise voice platform, because it was designed to be configured rather than rebuilt.