Clarison - Intelligent Transcription

Project Overview

Clarison is a full stack intelligent transcription platform built to deliver high accuracy speech-to-text conversion with context aware refinement. Developed from November to December 2025, the platform combines on-premise AI models with cloud-based LLM processing to provide sector-specific transcription services for medical, legal, and other professional domains.

Key Features

• Low Latency Speech Recognition: Integrated on-premise Ollama Whisper for fast, reliable speech-to-text conversion without cloud dependencies
• Multi Speaker Diarization: Python microservice that accurately identifies and separates different speakers in conversations and meetings
• Context Aware Refinement: Dynamic multi-stage processing pipeline powered by Gemini LLM that refines transcripts based on sector-specific requirements
• Accuracy Rate: Achieves over 90% accuracy in generating professional tone output tailored to medical, legal, and business contexts
• Sector Specific Processing: Customizable prompts and processing pipelines for different professional sectors with specialized terminology
• Architecture: Built with Node.js, Express.js, TypeScript, and React for a seamless user experience

Technical Architecture

The platform employs a sophisticated multi-stage architecture that combines real-time transcription with intelligent post-processing. Audio input is first processed through the Ollama Whisper model for initial transcription, then passed through a Python-based diarization microservice to identify speakers. Finally, the transcript undergoes context-aware refinement using Gemini LLM with sector-specific prompts to ensure professional-grade output quality.

Technology Stack

Frontend: React, TypeScript
Backend: Node.js, Express.js, Python
AI/ML: Ollama Whisper (on-premise), Gemini LLM
Processing: Multi-speaker diarization microservice

For more info on the code

GitHub – pritamsap/Clarison-IntelligentTranscription