AI Wednesdays: The One About Constraints, Speculations & Advanced RAG
More about the Sharing
Anirudh Shrinivason will take us "Beyond the Basics: Advancing Retrieval-Augmented Generation". The talk will cover advanced techniques in Retrieval-Augmented Generation (RAG) to enhance accuracy, efficiency, and scalability in AI-driven applications. We’ll dive into optimizing retrieval pipelines with hybrid search and embeddings, improving context relevance through chunking and re-ranking and some trade offs and considerations when building RAG pipelines.
James Teo & Nicole Ren will share about their innovative approaches to Free-Text-Anonymisation (FTA) using Small Language Models (SLM) and Constrained Decoding. Learn about their robust anonymisation solutions that harness the power of small but mighty SLMs, coupled with constrained decoding techniques. From enhancing performance to enabling customised detection, this talk will showcase how SLMs are punching above their weight in the critical field of text privacy.
Wilson Ng (Data Scientist, GovTech) from Govtech's AI Practice would be presenting on speculative decoding. Learn more about how speculative decoding is utilized to speed up large language models for accelerated inference and discover how fine-tuning models supported by speculative decoding can enhance their performance. (Technical Level: 300)
Sharings will start at 3pm!
More about the Speakers
Anirudh Shrinivason is a Software/AI Engineer with expertise in data platforms, applied ML, and GenAI solutions. He began as a data engineer at Grab and now works at Cohere as Member of Technical Staff in the applied ML team. His experience spans building scalable data platforms, enhancing data discovery, and developing GenAI frameworks and solutions, along with leading ML initiatives in multi-agent systems,RAG and applied LLMs for various other use cases.
James Teo is a Data Engineer from the Data Engineering Practice at GovTech where he works on exploring synergies between AI and Data Engineering such as Data Discovery with Metadata Knowledge Graphs and Data Contracts.
Nicole Ren is also a Data Engineer from the Data Engineering Practice at GovTech where she explores emerging Privacy Enhancing Technologies (PETs) such as Synthetic Data Generation and Homomorphic Encryption.
Wilson Ng is a Data Scientist in the AI Practice at GovTech, specializing in language models and areas such as fine-tuning, retrieval-augmented generation (RAG), and AI agents.