On top of the quantitative data capture proposed above, you may want to gently ask some users in those villages to let you observe them.
You could for instance just set up a camera to record the room they use and leave them alone with the web portal. Watching their reactions might help to tell you when they're frustrated, not finding what they want, and when they have a "A-ha" moment indicating that they found what they were looking for. You may also see when they ask for help to a peer, and whether the peer is able to help them or if they tell them to do something wrong or sub-optimal. This is useful if you're worried that your presence might intimidate the users or cause them to systematically ask you instead of trying.
You could also apply a think-aloud protocol, basically sitting next to the users and having them say aloud their thoughts when using the portal. This is useful if you want more quick details about where users struggle. Applying this protocol with three or four users should already give you insights on your portal's most prominent issues.
If you're really unfamiliar with user-centered evaluation, such methods might give you more visible and actionable data than just numbers. If you know only time gaps and sequences of actions, you still have to analyse which correspond to issues with the website and which are normal; maybe it's not so useful early in your project. It's great to have exact data on what pages users visit, which pages they stay the longest on, etc. when you have a lot of users and a large website. You can then detect the unusually long or short pages, or pages where users leave abruptly, and speculate about what's going on on those pages. Once you want to investigate a page or a task using a bunch of real users.