AI Researcher

A FinTech start-up is looking for a AI Researcher with deep model infrastructure knowledge to work on some complex issues regarding the trading industry.They are transforming this space with a CEO at the helm who has spent years and years in the industry himself before taking on the Tech start-up world in order to solve some of the common downfalls in the industry from a technology perspective. Backed by one of the biggest energy traders in the world, this business has stability but also the resources to go forwards and expand.You get to work up close with founders and the founding engineering team with significant ownership over how problems are solved, in fact, you'd be their first hire in this and working closely with the CEO.This is not an AI wrapper or prompt engineering role. They’re looking for someone who has genuinely worked with ML models and researched into what the current trends are from either a scientist or researcher perspective.What you’ll be working onDeveloping and deploying LLM-powered systems using PyTorchWorking directly with the CEO on synthesising training corporaHandling datasets and mainly dataset regularisationBeing able to showcase knowledge into linear attention mechanisms (e.g. gated delta nets)Emphasising diffusion speculatorsDealing with MoE routing collapseWhat we're looking forYou’ll ideally have hands on experience with several of the following:PyTorchTransformer-based models such as BERT, RoBERTa, Llama, etc.LLM fine-tuningLoRA / QLoRA / PEFTEmbedding modelsReranking / cross-encoder modelsSynthesising training corporaDataset regularisationGated Delta Nets or other Linear Attention mechanism knowledgeSpeculative decodingRouting or network collapse - MoEPlease note that academic, research or R&D experience is absolutely relevant.Experience with RAG is useful, but this isn't a role for someone whose experience is primarily connecting APIs together with LangChain or similar frameworks.I am particularly interested in people who understand what is happening underneath the RAG stack, how embeddings are trained and evaluated, how retrieval works, how re-rankers work, how training data is created and how models can be improved through experimentation.The role is 4 days a week hybrid in London and is paying up to the £110k + bonus mark on the top of the band.If interested, please apply or email me directly at tomasz@virtuetech.io