Alexandros Lazaridis

Staff Data Engineer in the DNA group, building LLM-powered
conversational insights on AWS/Bedrock and MS Azure, with openAI and
Claude, and on-prem models to understand customer needs and drive
self-service.
The project is focusing on Conversational Insights for understanding
and analysing what the customer needs by using Large Language
Models (LLMs), analysing our chatbot and voicebot data.
Tech Stack
Cloud/Data: AWS (S3, Glue, Lambda, Step Functions, Athena,
Bedrock), QuickSight (SPICE), Python, SQL
LLMs: Claude (Anthropic), Azure OpenAI, OpenAI, on-prem models,
fine-tuned/adapted models, prompt engineering
ML/Analytics: Embeddings, PCA/UMAP, HDBSCAN/Louvain, eval
harnesses
Governance: PII redaction, access controls, auditability, cost/latency
monitoring

Led 2 cross-functional teams (engineering, ML, analytics) over 3+ years across three products: Swisscom Voice Assistant on TV Box, VoiceBot/Chatbot for customer self-service, and Conversational Insights for customer-needs understanding.
• Defined product vision, strategy and roadmap, aligning with senior stakeholders, market trends, and customer needs.
• Owned backlog prioritization; delivered high-value features across speech recognition, NLU, retrieval (FAQ), machine translation, and LLM-based insights.
• Implemented metrics & dashboards driving data-informed decisions across product, engineering, and business.
• Partnered with engineering on technical feasibility, scope, and risk — translating between business and technical audiences.
• Proactively identified and mitigated delivery risks across timelines and dependencies.
• Mentored junior team members on Agile/Scrum, product ownership, and stakeholder management.

• Main focus of the Speech Processing R&D Group:

- Automatic Speech Recognition (ASR) on Swiss Languages and Dialects,

- Language Identification (LID),

- Speaker Identification/Verification (SI/SV),

- Emotion Identification & Sentiment Analysis (SER),

- Text to Speech Synthesis (TTS),

- Data Selection and Active Learning,

- Natural Language Processing & Understanding (NLP/NLU),

- Neural Machine Translation (NMT),

- Swiss German Dialects Text Normalization

• Leading R&D Projects:

- Swiss German Automatic Speech Recognition (in collaboration with Microsoft Research Center in Munich)

- DAHL: Domain Adaptation via Hierarchical Lexicons (In collaboration with Idiap Research Institute and École Polytechnique Fédérale de Lausanne (EPFL))

- Building A High German to Swiss German Phonetic Dictionary (In collaboration with the University of Zurich (UZH))

• Supervising and co-Supervising Master Thesis and PhD Projects

- In collaboration with Idiap Research Institute and École Polytechnique Fédérale de Lausanne (EPFL) - End-to-end Speaker Recognition from Speech, Self-attention for speech emotion recognition, End-to-end Language Recognition from Speech using Bottleneck Features, Data Selection for Speech Recognition Training, Domain Adaptation in End-to-End Speech Recognition, Key-word Search/Spotting using Convolutional Neural Networks, Semi-Supervised Data Acquisition and Explorations

- In collaboration with La Haute école spécialisée de Suisse occidentale de Fribourg (HEFR) - Swiss German Normalization with Neural Networks, Swiss German dialects Text Normalization using Neural Network Machine Translation-based Approaches and Word Embeddings

- In collaboration with University of Zurich - Fairness in AI.

• Post-doctoral Researcher in speech processing
• Supervised by Dr
• Philip N
• Garner
• Working on SIWIS Swiss NSF project
• The SIWIS (Spoken Interaction With Interpretation in Switzerland) project aims to investigate spoken language translation, where translation carries not only spoken content, but also prosody/emphasis in speech
• In the framework of SIWIS, worked on prosody modelling (phone duration and intonation modelling), regional accent identification using acoustic and prosodic features, stress and accent modelling, and statistical parametric speech synthesis using hidden Markov model (HMM) and deep neural network (DNN) techniques in English, German and French languages
• Research interests include signal processing, prosody modelling, speech synthesis, emotional/affective speech processing, data mining & machine learning, voice conversion.

• Research Associate in speech processing working on several projects: on speech synthesis and prosody modelling on the SIWIS (Swiss NSF) project, on multilingual automatic speech recognition on the SUMMA (EU-H2020) project and on automatic speech recognition for Swiss German dialects on Swisscom project.

• Course: Digital Design I.

• Courses: Digital Signal Processing, Informatics I and II.

• Continuing working on prosody modelling of neutral and emotional speech
• In parallel working on neutral and emotional speech synthesis using statistical parametric speech synthesis approaches.

• Course: Speech Technology.

• Course: Introduction to Computers.