How Atlas CI interprets your request
When you send a prompt, Atlas CI breaks it down into a step-by-step plan for research and analysis. It identifies the key entities in your request (such as drugs, clinical trials, companies, and indications), works out what you are asking for (a search, comparison, summary, or timeline), and finds the records that match.
If your request is ambiguous, Atlas CI picks the most likely interpretation based on context and its own training. It will then run the analysis, and show a play-by-play of how it approached your question.
How it works through a task
Atlas CI creates a research and analysis plan based on its interpretation of the prompt, the output will consist of text in the chat window and downloadable artifacts.
For a straightforward lookup, Atlas CI answers in a minimal number of steps. For broader questions — comparisons, summaries across many records, or landscape and trend questions — it breaks the request into smaller research tasks, works through them in order and will then draw on specialist tools to create the downloadable outputs.
At each step it will translate your prompt into a research plan to look up the relevant data in Norstella’s curated enterprise dataset
If more than one source is relevant, i.e. the request synthesises commercial, clinical and regulatory data, it queries each area in turn and brings the results together into a single answer.
It starts broad and drills into detail only where needed. For a company question, for example, it looks at a high-level pipeline summary first, then takes a closer look at a specific indication.
It reads the records relevant to your question, not the entire dataset.
If Atlas CI cannot find a clear match in the data, it may either expand its search or find alternative routes to obtain the information. Should it exhaust all these options, it will fallback on its own reasoning (inference).
If the agent is unable to identify the correct records within the dataset, this may increase the risk of inaccuracy within the output.
Asking for clarification
Atlas CI’s research and analysis plan will be based on what it perceives as the most likely interpretation of your query and its play-by-play will provide indications of the assumptions it has made.
When the query is ambiguous, e.g. when a request could reasonably be read in more than one way. Atlas will share what it found on its most likely interpretation, and you may clarify in a subsequent prompt to help Atlas CI update and rerun the research and analysis with the corrected understanding.
To reduce misinterpretation, it is recommended to hone the prompt help the correct interpretation. Include therapeutic area, drug names, trial details, company names, time ranges, and how you would like the agent to generate the output.
What Atlas CI does on its own
Atlas CI will interpret and answer your query in the chat interface.
It will automatically:
Search its connected dataset and retrieve matching records.
Summarise and structure the results.
Refine a search or find an alterative route to the relevant information if the first attempt is unsuccessful.
Choose the most likely interpretation and report the scope and assumptions it used.
It will not (currently):
Run ongoing or scheduled monitoring (there are no recurring or automated jobs).
Automatically export or share outputs to external files or parties on your behalf.
What Atlas CI generates
Atlas CI’s response to your query will include both a written answer with analysis and downloadable artifacts.
Atlas CI produces core CI artifacts, each built from records in the connected dataset:
Landscape summaries — development candidates across stages, from preclinical through to marketed products.
Clinical trial timelines — key events mapped across the trial lifecycle.
Key Events & Catalysts — a summary of key events and catalysts e.g. Drug approval, data readouts etc for the Drugs/Companies of interest as well as the impact of these events and catalysts.
Approval timelines — a summary of approval dates or estimated approval dates for relevant drugs of interest
Atlas CI can also generate more standard visuals including but not limited to bar charts, pie charts and simple tables.
Each artifact is delivered in the chat and can be downloaded into an editable ppt.
Currently Atlas CI generates documents in a fixed format. If you need the output in a tailored format, it is possible to download the file and copy and paste elements into your own templates.
Document content reflects what is recorded in the dataset at the time of generation based on the query you have submitted to the agent.
Default filtering behaviors in the agent
In order to return you with the most relevant drugs the agent applies a number of filters when retrieving data. These filters include:
A recency filter – The agent will only look at drugs, trials or events that have been updated in the past 5 years. This means older drugs that have had no new events or updates in this time period may be excluded.
A NME/NDA filter – The agent will only look at drugs that are NMEs, NDAs or New Derivatives, meaning generics and biosimilars will be excluded by default.
An active filter – The agent will only look at drugs that are in active development or marketed, drugs that have been abandoned or discontinued will be excluded by default.
These filters have been applied to ensure that the agent returns the most relevant results to you, however all these default parameters can be overridden within your prompt. If you wanted to look at abandoned drugs for example you can ask the agent to return drugs that have been abandoned and this will overwrite the default filter applied by the agent. These filters will only be applied when the agent has not received any other instruction to request otherwise.