Development of AI models is moving fast with new models built for different purposes being introduced constantly. AI and a 'Large Language Model' or LLMs have become interchangeable terms, but we deploy multiple different best in class models to deliver our platform.

Countless companies are investing huge sums of money extremely quickly to bring better, smarter models into the world. Our job is to evaluate these models for efficacy, security and applicability - and if deemed appropriate then integrate them into GovWorx.

We do not train our own models from scratch - not only is it cost prohibitive but unlikely to yield better results than what some of the largest companies in the world have achieved.

Our transcription pipeline is 100% within our network running on machines we control without any internet connectivity. This pipeline is responsible for transcription, speaker identification and tagging segments as having PII.

Batch Transcription

Batch transcription is run for all uploaded media (Calls, Radio Transmission, BWC, etc). The transcription model is OpenAIs Whisper which is fully open sourced with no data moving off GovWorx servers.

Streaming Transcription

Streaming transcription is used for CommsCoach Assist. The model runs fully inside GovWorx servers using NVIDIA Riva and Parakeet Model.

We have evaluated many models, but GPT is the best performing with a track record of the most cutting edge new releases. Balancing this with keeping our customers data secure. We have ensured this model