Mindloom is an intelligent knowledge organization platform built for technical professionals who accumulate documentation, articles, tutorials, and notes from dozens of sources with no common index. Rather than relying on memory or manual tagging to rediscover content, Mindloom replaces that effort entirely: every piece of content you ingest is automatically classified into one of eight technical categories, enriched with extracted keywords, linked to semantically related items, and made immediately searchable by meaning — not just by matching words.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/No-Country-simulation/G9-LATAM-Team-58/llms.txt
Use this file to discover all available pages before exploring further.
The Problem Mindloom Solves
Anyone who works or studies in technology accumulates content across multiple sources, often in multiple languages, with no consistent organizational scheme. Finding something read three months ago — or noticing that two articles cover the same concept with different vocabulary — depends entirely on the memory of whoever saved them. Mindloom moves that organizational burden from human memory to a structured, queryable index that grows smarter with every piece of content added.Six Core Capabilities
Thematic Classification
Each item is automatically assigned to one of 8 technical categories using a trained classification model backed by a multilingual sentence encoder.
Keyword Extraction
The most relevant terms in each document are identified and stored alongside the content for filtering, display, and explainability.
Related Content
Semantic similarity between all indexed items is computed at ingestion time using 384-dimensional embeddings, enabling instant recommendations without a separate ranking pass.
Semantic Search
Search by meaning using
GET /search?mode=semantic. A query is vectorized by the Inference Service and Oracle Autonomous Database resolves the nearest neighbors with VECTOR_DISTANCE.Knowledge Map
Every document is projected to 2D coordinates (UMAP) at ingestion time. The Web layer renders the full corpus as an interactive scatter plot colored by category — including content added live.
Incremental Indexing
Each new document immediately enriches the index. Related-content recommendations and map coordinates are computed for every item at the moment it is ingested, with no batch reprocessing step.
The Eight Categories
Mindloom organizes all technical content into exactly eight categories. These categories emerged from the real distribution of technical content used to train the model — they are not an arbitrary taxonomy:| Category | Description |
|---|---|
| Backend | Server-side development, APIs, frameworks like Spring Boot |
| Frontend | Browser interfaces, React, CSS, component libraries |
| Móvil | iOS and Android development, cross-platform frameworks |
| Datos e IA | Data science, machine learning, AI tooling |
| DevOps y Cloud | CI/CD, containerization, cloud infrastructure |
| Bases de datos | Relational and non-relational databases, query languages |
| Seguridad | Application security, authentication, cryptography |
| Fundamentos | Computer science foundations, algorithms, networking |
probability score (0–1) alongside the category label so consumers can surface confidence information in the UI.
System Overview
Mindloom is composed of four layers that form a clear ingestion and retrieval pipeline: A user pastes an article in the Web interface. The API (Spring Boot, Java 25) receives the request and forwards only the body text to the Inference Service (FastAPI, Python 3.12), which returns the category, probability, keywords, explanation tokens, a 384-dimensional embedding, cluster ID, and 2D map coordinates — all from a single call toPOST /predict. The API then persists the full record to Oracle Autonomous Database, storing the embedding in a native VECTOR(384) column. From that point on, the new document is a candidate for related-content recommendations and appears in the knowledge map.
Explore the Docs
Quickstart
Get Mindloom running locally in under 5 minutes using Docker Compose.
Architecture
Deep dive into the four-layer design, ingestion flow, and Oracle Vector Search integration.
Core Features
Explore classification, semantic search, the knowledge map, and incremental indexing in detail.
API Reference
Full reference for every REST endpoint exposed by the Spring Boot API.