@Shubham_Joshi thanks for posting! Finding a low cost data source is a challenge in the medical industry. Those that have spent the time vetting the data. Have that security mote. I asked perplexity is there an LLM for drug to drug interaction. The answer is below. But this would be for a clinic that is not using a paid eRx service or a country that does now have a national prescription system, imo.
Yes—there are LLM-based systems for drug–drug interactions (DDIs), including specialized research models and general-purpose assistants. The important distinction is between a model that can discuss interactions and a system reliable enough to guide prescribing. Current evidence does not support treating a standalone LLM as the final authority on whether a drug combination is safe.[pubmed.ncbi.nlm.nih][pubmed.ncbi.nlm.nih]
What exists?
| Type |
Examples |
What it does |
| General-purpose LLMs |
ChatGPT, Claude, Gemini |
Answer interaction questions and explain possible mechanisms; these have been evaluated on clinical case summaries. [pubmed.ncbi.nlm.nih] |
| Specialized research systems |
DDI-LLM, DDI-JUDGE |
Predict interactions using approaches such as molecular-graph information, biomedical information, and task-specific prompting. These are research approaches, not equivalent to validated prescribing tools. [sciencedirect][mendeley] |
| Biomedical text models |
BioBERT-based systems |
Extract reported drug interactions from published text—a different task from deciding whether a particular patient can safely take two drugs. [pubmed.ncbi.nlm.nih] |
How reliable are they?
Results are promising but uneven. In one 2025 study, ChatGPT, Claude, and Gemini detected many interactions in pharmacovigilance case summaries, but they also reported interactions that were not present in negative-control cases. Correct answers across all interaction-related questions ranged from 33% to 68%. Those results apply to the particular models and cases tested—not every current version or clinical setting.[pubmed.ncbi.nlm.nih]
That highlights a crucial limitation: identifying a possible interaction is not the same as reliably confirming that there is no interaction, assessing its clinical significance, or choosing the right management. Reviews identify hallucinations, data quality, explainability, and prospective clinical validation as unresolved issues.[pubmed.ncbi.nlm.nih][pubmed.ncbi.nlm.nih]
What makes sense in practice?
For checking a real medication list, a prudent approach is to use an established drug-interaction reference with pharmacist or prescriber review, rather than rely on an LLM alone. An LLM can help explain the findings, but should not independently clear a combination as safe.[pubmed.ncbi.nlm.nih][nature]
If you are building software, a more defensible design is to have the LLM retrieve and explain documented interaction records from a maintained drug database, rather than generate interaction claims from memory. That addresses some—but not all—of the reliability and data-quality problems identified in the research.[pubmed.ncbi.nlm.nih]
Are you looking for a tool to check medications, or a model/API to build into an application?