Man, I got way too into this project. It started so simple in my head: upload a doc, ask a question, get a smart answer that actually made sense. No AI jargon, no BS. Just understanding. That was the vibe I was going for.
What I ended up building was "Questioning" – my first real dive into this thing called Retrieval-Augmented Generation (RAG). Basically, an AI that (in my head) really listened to what you gave it. But what I really ended up building was… well, a whole lot of learning about myself and what I didn't know.
The first lines of code? Disaster. One giant, messy Python file. Backend, embeddings, the whole shebang, just thrown in there. It worked, kind of. Until it really, really didn't.
Every time I tried to tweak something, it felt like the whole thing was about to collapse. That's when it hit me: writing a bit of code is miles away from actually building something. So, yeah, I nuked it and started over. Small, separate pieces. Trying to make some kind of sense of it all.
That second try? That felt like actually figuring something out for the first time.
Then came the rabbit holes. Annoying dependency errors. Weird LLM hallucinations. Token limits making my app just choke? Seriously frustrating. The first time it cut off mid-sentence, I was convinced I'd broken the universe. Nope, just hit some dumb limit.
And that's when I started to see how these AI things act up,sometimes they just… make things up. Answers were too short, then too vague, then just plain wrong. Hours spent tweaking, prompting, swapping models.
"Why the hell isn't this doing what I expect?"
"What's the least janky way to pull the right info?"
"Am I even on the right track here, or just doing the easiest thing I can think of?"
Plenty of nights I wanted to just walk away. Felt like I was in way over my head. But every time I fixed something, every time I screwed up and figured out why, it was like this little lightbulb going off about how these systems actually "think." And then, one day, it just… clicked. The answers started making sense. The thing felt… usable.
That moment – when I asked a question and it gave me an answer that was spot-on, pulled right from the document I uploaded – that was wild. Felt like I'd built something that could actually understand. And suddenly, all the chaos before? Totally worth it.
But here's the kick in the pants. While I was obsessing over this little project, wrestling with the nitty-gritty of RAG, the big dogs – ChatGPT, Google – were doing the exact same thing, but on a scale that makes my little "Questioning" look like a toy.
Think about it. My whole idea – upload your stuff, get smart answers based on it – that's just… a feature now. Custom GPTs, Google's Vertex AI… they've got this RAG thing baked right in, all fancy and polished.
And it makes you wonder, right? All that time, all that effort… was it just a waste? Like, the thing I was so proud of building is just table stakes now.
But deep down, I know it wasn't pointless. Yeah, "Questioning" probably isn't going to change the world. But building it? That changed me.
So, yeah, "Questioning" might not be some groundbreaking app . But it was my personal deep dive into how this stuff actually works. It was me trying to figure something out, and in the process, figuring out a bit more about myself. And honestly? That doesn't feel pointless at all.
It feels like the entire point.