brianletort.ai

My Stack & Lab

The tools, frameworks, and hardware I use daily to build agentic AI systems. I love writing code—early mornings, weekends, whenever the ideas are flowing.

Home Lab Philosophy

A four-GPU home lab on the Mississippi Gulf Coast — RTX 6000 (96GB), DGX Spark (128GB), RTX 5090, and RTX 4090 — running Qwen, Gemma 4, and a rotating set of open-weight models for fine-tuning, local inference, and rapid prototyping. When you’re iterating on multi-agent architectures and context engineering, fast local feedback loops are everything. No cloud costs, no latency, no wait for a GPU to free up.

Home GPU Lab

  • NVIDIA RTX 6000 (96GB)

    Workhorse for larger-model fine-tuning and long-context experiments

  • NVIDIA DGX Spark (128GB)

    Unified-memory AI appliance for agentic development and local evals

  • NVIDIA RTX 5090 (32GB)

    Fast inference and daily local model work

  • NVIDIA RTX 4090 (16GB)

    Secondary rig for parallel experiments and quantization tests

  • Apple M5 Max MacBook Pro

    Daily driver — editor, tests, large MLX inference on the go

  • Local NAS

    Terabytes of training data, eval sets, and knowledge bases

Development Environment

  • Cursor

    AI-native code editor—my daily driver

  • VS Code

    Notebooks and specific workflows

  • Warp / iTerm

    Modern terminal for heavy CLI work

  • GitHub

    Version control and collaboration

AI Coding Assistants

  • Claude Sonnet / Opus

    Primary AI assistant for complex reasoning

  • OpenAI Codex / GPT-4

    Code generation and analysis

  • GitHub Copilot

    Inline completions and suggestions

  • Local LLMs

    Ollama, vLLM for private experimentation

Cloud & Infrastructure

  • Azure

    Enterprise cloud and AI services

  • Vercel

    Frontend deployment and edge functions

  • Supabase

    Postgres + real-time for rapid prototyping

  • n8n

    Workflow automation and agent orchestration

AI/ML Platforms

  • OpenAI API

    GPT-4, embeddings, and function calling

  • Anthropic Claude

    Complex reasoning and long-context tasks

  • Azure OpenAI

    Enterprise deployments with compliance

  • Hugging Face

    Model hub and specialized models

RAG & Agent Frameworks

  • LlamaIndex

    Data framework for RAG applications

  • LangChain / LangGraph

    LLM orchestration and multi-agent flows

  • Semantic Kernel

    Microsoft's AI orchestration SDK

  • DSPy

    Programmatic prompt optimization

Vector Databases

  • Weaviate

    Open-source vector search with hybrid capabilities

  • pgvector

    Vector search in Postgres—simple and effective

  • Pinecone

    Managed vector database for production

  • Chroma

    Lightweight local vector store for prototyping

Python & Deep Learning

  • PyTorch

    Primary framework for model development

  • Transformers (HF)

    State-of-the-art NLP models

  • scikit-learn

    Classical ML and preprocessing

  • Weights & Biases

    Experiment tracking and observability

Frontend & Web

  • Next.js

    React framework—App Router for everything

  • TypeScript

    Type safety for complex applications

  • Tailwind CSS

    Utility-first styling

  • shadcn/ui

    Beautiful, accessible component library

  • Framer Motion

    Smooth animations and micro-interactions

Currently Experimenting With

12-15 Agent RAG Pipelines

Production

Specialized agents for retrieval optimization at terabyte scale

Context Engineering Techniques

Research

Memory optimization to overcome attention dilution

Plan-Act-Learn Pipelines

Prototyping

Self-learning ETL that adapts autonomously

Real-Time Hallucination Detection

Development

Quality feedback loops with continuous learning

Local Fine-Tuning Workflows

Development

LoRA and QLoRA on consumer GPUs for domain adaptation

Headless SaaS Patterns

Exploration

Agent-first API design for the post-GUI era

Research Focus Areas
  • Multi-agent parallel architectures for retrieval optimization
  • Recency bias and attention dilution mitigation
  • Reinforcement learning for RAG response quality
  • Forward-thinking entity-linking for feature spaces
  • Autonomous data pipeline adaptation
  • Agent-to-agent communication protocols

Open Source & Community

I'm a huge believer in open source. Most of my stack is built on open frameworks—LlamaIndex, LangChain, Semantic Kernel, Weaviate, pgvector. These communities move faster than any single company. I love building with Next.js, shadcn, and the modern JavaScript ecosystem. When the ideas are flowing, I'm writing code—whether it's tweaking RAG pipelines or fine-tuning models on my local GPUs.