Oector.

The fastest way to answer security and compliance requests

Comment

Inspiration

I have seen security teams lose a lot of time having to answer the same questions from different clients and vendors. Most still use the same old excel sheet/knowledge library to find the questions and answer pairs. These are often outdated, or fragmented and need a lot of manual effort to scan through. So, I thought what is using AI it was possible to automate this task.

What it does

So basically Oector is a AI powered system for Trust and Compliance operations streamlining the hard manual labor. It provides following features:

  1. Helps import existing Knowledge bases, existing documents & policies etc.
  2. Detect duplicate, conflicting, and outdated knowledge entries.
  3. Helps automatically fillup security questionnaires with grounded truth and citation, from existing knowledge
  4. Also maps questions to their respective frameworks like: NIST 800-53 and OWASP ASVS.
  5. Constantly Looks for Compliance Policy updates and helps update the Knowledge base regularly

How we built it

It is built on FastAPI for the backend that does the heavy lifting, PostgresSQL as the database along with pgvector for semantic retrieval, Redis/Celery for background tasks, a local LLM / Online LLM provider for embedding and answering.

Challenges we ran into

The main issue was being able to make the AI output useful and workable. Another was being able to create a pipeline that tracks back the citations & scores the answers while also allowing quick fixes. Other task was to try to make is efficient enough such that it doesn't go in O(n2) complexity. when embedding and semantically searching and flagging question answer pairs. Also trying to be able to import data from different source through established ways.

Accomplishments that we're proud of

  1. Firstly to be able to complete this like a end-to-end system not just another wrapper,
  2. Have a periodically scanning infrastructure that scans for new compliances changes across all areas, and also effected entries

What we learned

Building a scalable system, that could accommodate large dataset and process is parallelly is really difficult, being able to think clearly and brainstorm correct path toward the goal is really important. Also small things and features can also make a huge impact of the overall experience of the solution

What's next for Oector.

Firstly, would be more deeper integrations, into more knowledge bases, and document storages etc. Other would be adding more analytics features, and this is still not as efficient as I would like to be so work on the efficiency too

Share this project:

Updates