Lohith Konathala

Research Engineer @Neuphonic

London, GB

Developed speech-language models combining LLMs with neural audio codecs, trained using large-scale proprietary datasets. Experience with post-training alignment techniques including Direct Preference Optimisation (DPO), weight-interpolation, and reward-guided decoding to improve model robustness.Experience quantising and deploying SpeechLMs on-device using Ollama for real-time inference on edge-devices.