2026-07-21 –, Room 2.41 (First Floor, Turing)
Recent advances in AI have shifted industries’ attention toward integrating LLM-based systems. Even though LLMs can solve a wide range of business problems, they came with a significant complexity overhead. At same time, many real-world business applications involve well-defined objectives, predictable inputs, and clear evaluation criteria.
Today, we are increasingly seeing a default pattern: for almost any NLP use case, teams prompt GPT-like models and pay the bill at the end of the month. However, this approach often introduces unnecessary complexity, costs, and operational risk. Many business and research problems exist in constrained environments that can be solved with simpler techniques, achieving the same or higher success rates.
This talk defends that fine-tuned BERT-based models remain a strong and often superior choice for targeted business use cases that require NLP-based solutions. I propose to present a real, in-production use case where a simple transformer-based classifier demonstrates a more favourable performance-cost trade-off than LLM-based approaches, driven by lower latency, reduced operational complexity, easier fine-tuning, and significantly lower maintenance costs.
The goal of this presentation is not to reject LLMs, but to promote a pragmatic, outcome-driven approach to NLP, where “boring” solutions often deliver the most value.
Recently, LLMs have become the default answer to almost every NLP problem. Need classification? Prompt an LLM. Need tagging? Prompt an LLM. Need summarization? Prompt an LLM. It is undeniable that it works, but often at a cost that businesses quietly absorb without always seeing proportional value.
This talk takes a step back and asks a simple question: Are we solving business problems, or just using the most trendy tools available?
Through a real production case study, I’ll show how a fine-tuned BERT-based classifier solved a well-defined business problem more efficiently than an LLM-based alternative, with lower latency, reduced operational complexity, easier maintenance, and significantly lower cost. In constrained environments with standardized inputs and clear evaluation criteria, simpler transformer models can still be the most effective solution.
Importantly, this is not an anti-LLM talk. LLMs play a valuable role in the workflow, from data exploration and labelling support to analytics such error analysis. But when it comes to production inference for well-defined tasks, smaller task-specific models often deliver the best performance-to-cost ratio.
If AI is going to succeed in business, it won’t be because models are bigger, it will be because solutions deliver measurable impact without unsustainable cost or complexity. As data scientists and AI engineers, designing those efficient trade-offs is part of our responsibility.
Sometimes, the most impactful AI solution is also the most boring one.
Senior AI Engineer with 7+ years of experience architecting and deploying end-to-end ML solutions at scale. Specialized in NLP, Generative AI (LLM, RAG), Vector Search, and MLOps.