At Sustainable AI Solutions, I built live speaker re-identification using SpeechBrain ECAPA-TDNN embeddings and cosine-similarity matching. I also added local summarisation with Qwen3-4B through llama.cpp.
I integrated Azure AI Speech transcription with speaker diarisation and built summarisation on Azure Foundry. I deployed the work to Azure Container Apps from Azure Container Registry.
For my Master thesis at Ericsson, in collaboration with Ericsson Research - Trustworthy AI, I developed a multi-agent chatbot that uses LLMs to generate SQL queries and explain reinforcement learning agent actions. I also built a RAG pipeline for retrieving and summarising insights from telecom datasets.
At Immersion, I built and maintained data pipelines for 60+ users, improving resource utilization by 72%. I also analysed more than 1,200 projects, developed regression models to forecast rendering time and delivery timelines, and created dashboards and automated KPI reports.
