Structural elucidation from Nuclear Magnetic Resonance (NMR) data has traditionally been approached as a modeling problem: training neural networks to map spectra to molecular structures. However, recent research shows that reframing this task as an agentic search problem, guided by large language models (LLMs), yields superior results without specific training. This approach not only transforms spectroscopy but establishes a replicable paradigm in enterprise software development: instead of building monolithic systems, we orchestrate autonomous agents that interact with specialized tools, validations, and knowledge bases.
In the NMR domain, the agentic system uses a frozen LLM (no fine-tuning) that, through step-by-step instructions mirroring a chemist's reasoning, queries chemical shift tables, executes processing tools, and validates hypotheses. On the Alberts dataset, the agent achieves 71% top-1 accuracy, surpassing the 66% of graduate students. On the van Bramer and AstraZeneca datasets, it reaches 80% and 20% respectively, outperforming deep learning models trained on thousands of simulated spectra. This result shows that the key is not modeling the direct transformation, but structuring discovery as a search constrained by expert knowledge.
This philosophy has a direct parallel in the software industry. At Q2BSTUDIO, we develop agentic AI solutions that, instead of relying on closed models, deploy agents that interact with APIs, databases, and legacy systems. Just as the NMR agent queries tables and tools, our business agents can access heterogeneous data sources, apply validation rules, and perform complex tasks autonomously. This architecture is ideal for custom software that requires integration across multiple platforms and a high degree of personalization.
The analogy extends to other services we offer. In cybersecurity, an AI agent can orchestrate scans, correlate events, and execute automatic responses, outperforming solutions based solely on detection models. Similarly, in cloud environments (AWS/Azure), agents can manage infrastructure, optimize costs, and enforce security policies without human intervention. Even in Business Intelligence (Power BI), an agent can explore data, generate reports, and answer natural language questions, combining statistical processing with domain knowledge.
The success of the agentic approach in NMR rests on three pillars that we apply to every project: first, a curated environment with specific tools (tables, scripts, validators); second, an LLM that acts as a reasoning engine but not a black box; and third, instructions that reflect an expert's cognitive process. In our automation developments, we apply the same principle: we design workflows where agents use continuous integration tools, data validation, and notifications, all orchestrated by an LLM that decides next steps.
For companies seeking digital transformation, this model offers tangible advantages: reduced training costs, dynamic updating (adding new tools or data is incorporated without retraining models), and transparency (each decision can be traced to the prompt and the queried tool). Q2BSTUDIO implements these architectures in sectors such as logistics, healthcare, and finance, where accuracy and adaptability are critical.
The future of AI lies not in ever-larger models, but in systems that know when and how to use specialized tools. NMR elucidation is just one example of how shifting from modeling to agentic search can yield significant advances. We invite organizations to explore this approach with us, whether for spectroscopic analysis, process automation, or any other challenge where artificial intelligence must dialogue with expert knowledge.
