Why can't Latin America build great nations?
Before answering why Latin America cannot build great nations, we need to define: what are great nations?
They are those that have managed to tra...
8 min
Read moreHi, I'm Fredy Rivera, a Full-Stack Developer and AI-Developer!
Tech Stack
React
Tailwind CSS
PostgreSQL
Vercel
GitHub
Git
Python
FastAPI
Redis
OpenAI
Anthropic
Stripe
Paypal
Supabase
DigitalOcean
AWS (EC2)
React
Tailwind CSS
PostgreSQL
Vercel
GitHub
Git
Python
FastAPI
Redis
OpenAI
Anthropic
Stripe
Paypal
Supabase
DigitalOcean
AWS (EC2)
My Projects
Things I've built
Web budgeting application
A budget management platform for construction companies
PythonFastAPITailwind-CSSHTMLPostgreSQLSupabaseVercel
Aquiles-RAG: High-Performance Vector Search System
Production-ready Retrieval-Augmented Generation (RAG) solution with multiple vector database support
PythonFastAPIRedisQdrantPostgreSQLRAGVector SearchHNSWAI
AtlasServer-Core: Self-Hosted Application Server
Fast deploy. No cloud. Just code. A self-hosted server platform with AI-powered deployment capabilities
PythonFlaskFastAPIDjangoCLISelf-HostedAIOllamaDevOpsNgrok
LLaDA-from-scratch: Diffusion Language Model Implementation
Building LLaDA, a diffusion-based language model that learns text distribution through progressive masking and reconstruction
PythonPyTorchHuggingFaceDiffusion ModelsNLPDeep LearningA100Research
Latest Articles
Recent thoughts and insights
Artificial Intelligence, War and Mass Surveillance
Artificial Intelligence is being deployed at scale in warfare. We have seen this in cases such as the use of Claude (Anthropic) in the capture of Vene...
6 min
Read more
Video
Fine-tuning Asclepio-8B and Qwen2.5-VL-3B: Medical Reasoning and Screenshot-to-Code with LoRA
Training specialized models doesn't require impossible GPUs or weeks of compute. Asclepio-8B learns clinical medical reasoning with 1.3M examples in 6.7 hours, reaching 76.9% accuracy. Qwen2.5-VL-3B converts UI screenshots into functional HTML/CSS with 94.6% accuracy in 5.5 hours. Both trained on L4 24GB using LoRA, demonstrating that specialized fine-tuning with smaller models can outperform giant general-purpose models. This post documents configurations, data pipelines, real metrics, and why fast iteration matters more than raw parameters.
fine-tuningLoRAmedical-ai+6
16 min
Read moreRequestScopedPipeline: Concurrent Inference in Diffusers without Race Conditions or Memory Duplication
Diffusers pipelines weren't designed for concurrency: calling pipe() simultaneously causes race conditions in schedulers, 'Already borrowed' errors in Rust tokenizers, or duplicates entire models in memory. My contribution (#12328) introduces RequestScopedPipeline, which solves these issues by creating lightweight per-request views, cloning only small mutable components, and adding automatic locks to tokenizers. Result: a server that handles multiple concurrent users without exploding GPU memory.
diffusersaiserver-async+3
9 min
Read morelambda-gateway: Building a Serverless Host Demo
Have you ever wondered how platforms like Vercel and AWS Lambda work under the hood? I built lambda-gateway, a serverless hosting demo using Docker, FastAPI, and Next.js to discover how serverless architectures work through hands-on experimentation.
lambdaserverlessdocker+4
15 min
Read moreMy First Blog Post
Welcome to my blog! This is my first article where I'm going to share my experiences as a Full-Stack developer. In this space I'll be writing about web development, artificial intelligence...
web developmentnext.jsreact+1
1 min
Read more