Skip to content
View Luiz-Frias's full-sized avatar

Block or report Luiz-Frias

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Luiz-Frias/README.md

$ whoami

I write Rust lints that deny unwrap() by day and quantum ethics papers by night.
If that doesn't make sense yet, keep reading. Or don't. I'm not your supervisor.
(Actually, I built the supervisor too. It's called SAGE.)


the tldr

I build the search engine, then the agent that uses it, then the orchestrator that deploys the agent, then I write a paper questioning whether the agent should have ethics derived from Pauli operators instead of vibes.

semantic-code-agx  →  agentic-codebase-navigator  →  SAGE  →  siglet-qubits  →  ???
     (find)                  (reason)                (build)     (reflect)       (repeat)

It's turtles all the way down except the turtles are hexagonal and they compile.


what I actually do

I build end-to-end systems where the output of one thing becomes the substrate for the next.

The common thread: everything I ship has a hexagonal architecture, strict dependency boundaries, and an almost pathological aversion to unwrap(). If the compiler can catch it, I refuse to find out at runtime.

Better yet:

Step 1: Teach machines to read code.

Step 2: Let them think about it.

Step 3: Make them work together.

Totally normal behavior.

🦀 semantic-code-agxfind code by meaning, not grep

A 12-crate Rust modular monolith. CLI-first semantic code search engine with local HNSW indexing, tree-sitter AST-aware splitting, multi-provider embeddings (cloud & local), and optional Milvus. Currently developing DFRR — a novel search algorithm. Details coming when the benchmarks do.

unsafe_code = "deny" · clippy::unwrap_used = "deny" · clippy::expect_used = "deny" · clippy::panic = "deny"

Yes, really. The whole workspace.

🐍 agentic-codebase-navigatorLLMs that actually run code until they're right

Hexagonal Python implementation of Recursive Language Models (RLM). LLMs iteratively write, execute, and refine Python code to answer questions. Multi-backend routing, relay pipeline DSL with type-safe state composition (>> for sequence, | for parallel, .when() for conditionals), Docker sandboxing, and tool calling. On PyPI and everything.

Built on research from MIT OASYS Lab — refactored from "it works" to "it works and I can explain the dependency graph."

🧠 SAGEthe thing that builds the thing

Supervisor Agent-Generated Engineering. A multi-agent code generation orchestrator with wave-based execution, fan-out/fan-in parallelization, quality gates at every milestone, and language-agnostic domain modules. You describe what you want. SAGE plans, assigns agents, enforces gates, and coordinates the build.

Dual-licensed AGPL-3.0 / Commercial because some things should be free and some things should pay rent.

📄 siglet-qubitswhat if ethics emerged from physics instead of philosophy

A framework for discovering ethical primitives through quantum-symbolic simulation instead of hardcoding them from human moral traditions. Siglets evolve under temporal dynamics; ethical structure emerges from stability, coherence, and spectral clustering — no semantic labels until post-hoc validation.

Epistemic humility as an engineering constraint. The paper exists. The rabbit hole goes deeper.


the pattern

         ┌─────────────────────────────────────────────┐
         │                                             │
         │   build it  →  search it  →  reason over it │
         │       ↑                           │         │
         │       └───── orchestrate ←────────┘         │
         │                                             │
         └─────────────────────────────────────────────┘
                    (repeat until mass adoption 
                     or mass confusion)

Every project feeds the next. The search engine makes the agent smarter. The agent makes the orchestrator useful. The orchestrator makes the next search engine faster to build. The paper asks if any of it should exist. I build it anyway.


how I build

[philosophy]
architecture   = "hexagonal, always"
error_handling = "typed envelopes, never strings"
validation     = "parse, don't validate"
testing        = "if it compiles but doesn't test, it doesn't exist"
scaling        = "modular monolith → microservices (when the data says so)"
decisions      = "empirical gates, not theoretical debates"
[tools]
languages  = ["Rust", "Python", "TypeScript"]
mantra     = "Rust for the kernel, Python for the glue, TS for the vibes"
ide        = "Cursor"
runners    = ["just", "mise", "bacon"]
testing    = ["nextest", "pytest", "proptest"]
packaging  = ["brew", "PyPI", "scoop", "winget"]

currently

  • 🔨 Building DFRR search algorithm (@ me, if you're curious)
  • 🌊 Expanding SAGE's domain modules
  • 🧪 Running siglet-qubit simulations that I can't fully explain to my friends
  • ☕ Drinking concerning amounts of coffee in Orlando

"We build the thing that builds the tools to build the thing — on auto ⚙️ and nitro 🚀"

luizfrias.com · threads · tiktok

If you've read this far, you're either recruiting me, competing with me, or equally unhinged. In all three cases: just keep swimming 🐠🎶

Pinned Loading

  1. semantic-code-agx semantic-code-agx Public

    CLI-first semantic code search engine — Rust modular monolith with hexagonal architecture, local HNSW indexing, multi-provider embeddings, and optional Milvus integration.

    Rust

  2. agentic-codebase-navigator agentic-codebase-navigator Public

    Hexagonal architecture implementation of Recursive Language Models (RLM)

    Python

  3. sage-system sage-system Public

    Context Engineering System developed by Luiz Frias

    Python 2 1