BLOOP Runs on Memory. Literally.

Most apps forget you the second you close the tab.

You come back the next day, same question, same confusion — and the app greets you like a stranger. Like nothing ever happened.

BLOOP doesn't do that.

BLOOP remembers that you've been stuck on thermodynamics for three days. It remembers how you like things explained. It picks up where you left off — because it actually kept track.

That's the whole idea. And Actian VectorAI DB is what made it real.

First — What Even Is BLOOP?

BLOOP is an AI learning assistant for STEM students.

Upload your notes. Ask questions. Get answers that make sense. Get video explanations when words aren't enough. Have the system actually remember you across sessions instead of treating you like a stranger every time.

Simple pitch.

Here's what it actually required us to build:

A system that reads and understands your documents.
A memory layer that knows you — not just your last message.
A video generation pipeline that produces animated explainers.
And somehow, make all of it searchable and retrievable without it turning into a debugging nightmare.

This is where Actian came in.

The Problem Nobody Warns You About

When you start building an AI product, you think the hard part is the AI.

The prompts. The models. Getting it to stop hallucinating.

You are wrong.

The hard part is: what does the system actually know, and how does it find the right thing at the right moment?

Because BLOOP needed to retrieve three completely different kinds of things:

Your documents — the PDFs and notes you upload. Your memory — the context about you that should persist across sessions. Your video artifacts — the animated explainers BLOOP generates.

In a typical hackathon build, you'd make three separate systems for these, tape them together with hope and caffeine, and spend the last four hours wondering why the memory retrieval keeps pulling up someone else's document.

We did not want to do that.

What Actian VectorAI DB Actually Did For Us

Here's the thing about Actian that changed everything:

We didn't need three systems. We needed one.

Actian let us store documents, memories, and video artifacts all in the same place — just with different labels on them. Same search. Same filters. One coherent layer instead of three separate headaches.

Ask a question → Actian finds the right document chunks. Start a new session → Actian pulls the memories that matter about you. Generate a video → Actian makes even that searchable.

All from one place. All consistent. All without us manually routing between different databases depending on what we needed.

That sounds like a small thing. At 2 a.m. during a hackathon, it is not a small thing.

**The search actually understood what we were asking**

Actian supports hybrid search — which means it combines semantic search (finding things by meaning) with more traditional keyword-based search.

In practice: a student doesn't always phrase things perfectly. They don't search "thermodynamic entropy second law" — they type "why does heat always go to cold??"

Actian figures it out anyway. It finds the right chunks. The answer makes sense. The student doesn't know or care how it happened — they just got helped.

That's the whole product. That's what we were trying to build.

The memory filter is the thing we're most proud of

Here's the trap with memory: if you save everything, you retrieve garbage. If you save nothing, every session starts from zero and BLOOP feels like a goldfish.

We wanted the signal — the stuff that actually matters about a user.

So after every answer, BLOOP asks itself: "Was that worth remembering?"

Not "okay thanks." Not filler. The real stuff — learning preferences, recurring topics, the clarification that finally clicked.

If the answer is yes, it writes a distilled memory into Actian. Next session, that memory comes back alongside the document context. The student gets an answer that feels like it knows them.

Actian made this possible because we could store and retrieve those memories with the exact same infrastructure as everything else. No new system. No new logic. Just a different label.

It held up when we kept changing things

Hackathons are chaos. Requirements change. Someone has a new idea at midnight. You need to add something that wasn't in the plan.

We changed our chunking strategy mid-build → retrieval still worked. We added new memory types on the fly → retrieval still worked. We wired in image search at the last minute → retrieval still worked.

The manim video pipeline? Still a little unhinged, not going to lie. Some prompts break it and we're working on that.

But Actian? Never the problem. Every time we threw something new at it, it held.

The Weirdest Flex: We Made Video Searchable

BLOOP generates animated math and science explainers. You ask about eigenvalues, you get a rendered animation.

Cool. But how do you search a video?

You don't search the video. You search the text that describes it.

Every video BLOOP generates produces scene descriptions, narration scripts, timestamps. We fed all of that into Actian as text records. Suddenly the video is queryable — just like a document.

It's a weird way to think about it. It's also completely correct. And it only worked because Actian could hold all of it in one place without us needing to build anything special.

What We Actually Learned

We came into this hackathon thinking we were building an AI product.

We left knowing we were building a retrieval product that happens to use AI.

The quality of BLOOP's answers lives or dies on whether the right context shows up at the right moment. The memory, the documents, the artifacts — all of it has to be findable, accurate, and fast.

Actian VectorAI DB is the reason that part works. Hybrid search, smart filtering, one layer for everything — it took what would have been our biggest source of debugging pain and made it the most stable part of the whole system.

We're proud of BLOOP. We're genuinely grateful for Actian.

And we're still a little surprised it all held together.

Command Palette