Applied AI

Armenian Voice-Assistant SaaS

An Armenian-language voice platform combining speech recognition, retrieval over business knowledge, secure APIs, and call analytics.

A fine-tuned Whisper speech-to-text model and retrieval pipeline sit behind a FastAPI service, while a React dashboard exposes calls, transcripts, and intent analytics.

View source code
Role
Full-stack and ML engineering
Status
Product prototype
Project type
Applied AI
Technology
Python · FastAPI · Whisper · PostgreSQL · Vector database · React · Azure STT
System signals
Model serving · Retrieval pipeline · Call analytics

Engineering contribution

What I worked on

  1. Fine-tuned Whisper using an Armenian speech corpus and served inference through FastAPI.
  2. Built knowledge-base retrieval over a vector database.
  3. Connected the model pipeline to a React operations dashboard.