Inspiration
We consume an enormous amount of useful information through short-form content, especially Instagram Reels, but most of it disappears into an endless feed. You might watch a Reel explaining a great book, a negotiation technique, a programming concept, or an interesting idea and think, “I need to remember this.” A few weeks later, it's nearly impossible to find again. We wanted to turn that passive consumption into something much more useful: a personal, interconnected memory system. That idea became MemoReel, a way to save the knowledge hidden inside Reels and transform it into a personal knowledge graph that you can actually query and explore.
What it does
MemoReel is a browser extension that lets you save an Instagram Reel with a single click. Instead of simply bookmarking the Reel, MemoReel extracts the useful context surrounding it, including its content, caption, hashtags, links, important people, titles, and concepts. An AI pipeline then analyzes the Reel, categorizes the information, identifies important entities and concepts, and connects them to related knowledge in your personal graph. Over time, your saved content becomes an interconnected knowledge base rather than a collection of bookmarks. You can then ask questions such as: “Based off of all interview prep reels I have consumed so far, how should I negotiate during an interview?” “What books have I saved about improving my mindset?” “What concepts have I learned about building startups?” “Show me the Reels related to machine learning.” MemoReel retrieves the relevant knowledge from your graph and uses an LLM to synthesize an answer grounded in the content you've personally saved. We also integrate with Obsidian, allowing the generated knowledge graph to become something you can visually explore and navigate.
How we built it
MemoReel is built around a browser extension, an AI processing pipeline, graph storage, and retrieval-augmented generation. The browser extension adds a lightweight interface directly to Instagram. When a user clicks 'Memorise', the current Reel is sent to our backend for processing. Our backend then:
- Extracts and processes the Reel's available multimodal content and metadata.
- Uses AI to identify important concepts, entities, links and topics and categorizes the content into a structured knowledge representation and creates a JSON-based summary.
- The JSON is stored in Databricks and is used to create nodes and relationships for the user's knowledge graph.
- We used Fast MCP to extract information about the content (links, other resources, etc.) by visiting the websites, using the search functionality offered by MCPs, and extracted more information from the source itself to make the content saved on the knowledge graph richer.
- We have also employed LangGraph to query the knowledge graph.
- The JSON is used to create one .md file for every reel and then the generated .md file is used to create an interactive Knowledge Graph on Obsidian.
- Finally, when the user clicks 'Explore' on the chrome extension, the user can actually interact with their own knowledge graph on our interactive 3D UI and also query the data that has been accumulated so far in the knowledge graph.
Challenges we ran into
One of the biggest challenges was turning messy, unstructured short-form content into structured knowledge. A Reel isn't naturally a database record. Important information can be hidden across spoken content, captions, hashtags, links, names, and visual context. We had to design a pipeline capable of extracting the information that is actually useful for building relationships between pieces of knowledge. Another challenge was deciding what should become a graph node versus what should simply remain contextual information. We also had to think carefully about retrieval. A purely vector-based search can find semantically similar content, while a graph can reveal explicit relationships between concepts. Combining the two gives us a much richer foundation for answering questions. On the frontend side, we wanted MemoReel to feel lightweight and native to the browsing experience, which meant building a draggable, expandable interface that doesn't get in the user's way.
Accomplishments that we're proud of
We're proud that we went beyond simply building another bookmarking tool. MemoReel transforms something ephemeral, a Reel you might otherwise forget, into a persistent piece of your personal knowledge system. We're particularly proud of the end-to-end pipeline: Instagram Reel → AI extraction → structured knowledge → knowledge graph → retrieval → personalized answer We also built the system so that knowledge isn't isolated to individual saved Reels. As more content is added, relationships between concepts become increasingly valuable. The Obsidian integration was another important accomplishment because it gives users a tangible way to visually explore the connections their content creates.
What we learned
We learned that building a useful AI system isn't just about choosing a powerful LLM. The difficult and interesting part is designing the structure around the model, deciding what information to extract, how to represent it, how to connect it, and how to retrieve the right context later. We also learned how powerful the combination of knowledge graphs + vector search + LLMs can be. Each component solves a different part of the problem: Vector search helps find semantically relevant information. Knowledge graphs preserve explicit relationships and structure. LLMs turn that retrieved information into useful answers. Most importantly, we learned that the best personal AI systems shouldn't just remember information, they should help people make connections between things they've already learned.
What's next for MemoReel
Our next goal is to expand MemoReel beyond Instagram Reels. We want to make the same personal memory layer work across YouTube, TikTok, articles, podcasts, screenshots, notes, and other sources of information. We're also interested in making the knowledge graph increasingly personalized, learning which concepts are important to a user, identifying gaps in their knowledge, surfacing forgotten information at the right time, and proactively connecting new content to things they've already saved. Ultimately, we want MemoReel to become a personal second brain that grows naturally from the information you consume every day, turning scattered content into connected, searchable knowledge.
Log in or sign up for Devpost to join the conversation.