There I was, uploading all of our IRIS reports to NotebookLM, wishing it would do a better job of extracting facts and figures. Then I thought, “I wonder if anyone else is doing this… why don’t we just do it for them?” As we tossed the idea around, we realized the timing was right given our work at the time to overhaul our website and the constantly growing library of IRIS and Retina content.
That was earlier this year, and now I’m happy to introduce you to the Cyentia Research Assistant (or assistant, for short). The idea is pretty simple, especially in a time of growing familiarity with LLM assistants: ask a question in plain language and get an answer drawn from our reports, data, and analysis—with citations so you can trace every claim back to its source. (Please don’t flame us for using em-dashes, we were into them before they were cool only then to become uncool again.)
It’s designed for analysts who need specific answers quickly. Rather than searching for a report and reading through it, you can ask what you actually want to know: which sectors show the highest incident frequency, how attack patterns have shifted over the past three years, what the loss parameters look like for your industry. The assistant finds the relevant content and synthesizes an answer, or tells you directly if its library doesn’t cover it.
A few things worth knowing:
- The assistant answers only from our IRIS and Retina artifact library—it won’t fill gaps with general knowledge or speculation. This is intentional: every response is grounded in research we stand behind, and every claim is citable.
- Your access reflects your membership. The content you see is scoped to your subscription tier (free or paid) and, where applicable, your industry. Sector-specific content surfaces automatically—you don’t need to specify it.
- Sessions are fresh each time—the assistant doesn’t retain memory between conversations.
- If a response is incorrect or misses the mark, you can use the flag button to send feedback directly to our team.
For the nerds out there (you know who you are…), the assistant is a retrieval-augmented generation (RAG) system. A user types a question, it identifies the most relevant content available to the user’s subscription tier from the artifact library, passes that content to GPT-4o alongside strict instructions to answer only from what’s provided, and streams a response back with citations. If the library doesn’t contain enough to answer the question (based on thresholds we can tune over time), the assistant says so rather than speculating.
And something we’re particularly proud of: for data tables (currently part of paid Retina content), the system can handle both lookup queries (“show me the frequency parameters for mid-size firms in the finance sector”) and aggregation queries (“which ATT&CK techniques are associated with the most losses?”)—the latter using a structured query engine that runs SQL against the underlying data rather than trying to infer the answer from search results. Oh, and charts referenced in responses render inline (hi-res, spared no expense).
It’s not perfect—e.g. the kind of table-like charts we’re partial to don’t play well with semantic search, so we’re considering back-filling them in other formats for better processing—, but I’m happy with what it can do. We’re grateful to anyone who spends time reading our work, and we genuinely hope this makes it easier to find what you’re looking for.