← Complete project index

Embedded & Devices / Built

ESP32 Command LLM

A 580K-parameter transformer trained from scratch to turn English and Spanish commands into GPIO, PWM and servo actions entirely on an ESP32-S3.

STATUSBuilt
AREAEmbedded & Devices
STACK5 documented technologies
INTEGRATIONS1 connected / bounded services
01

OVERVIEW

What I built.

A 580K-parameter transformer trained from scratch to turn English and Spanish commands into GPIO, PWM and servo actions entirely on an ESP32-S3.

01

Trained a domain-specific 580K-parameter transformer from scratch for device-control commands.

02

Implemented on-device inference that maps English and Spanish sentences into GPIO, PWM and servo actions.

03

Designed the model around a constrained command DSL rather than general-purpose language generation.

02

SYSTEM ARCHITECTURE

How the system is divided.

01
Architecture boundary

Natural-language command -> tokenizer/model in PSRAM -> structured command -> GPIO/PWM/servo executor.

02
Architecture boundary

Inference runs entirely on the ESP32-S3 with no cloud API and no PC in the loop.

03
Architecture boundary

The project targets one known hardware profile instead of pretending to be universally portable.

03

NETWORK / DATA FLOW

What talks to what.

04

TECHNOLOGY STACK

The technical surface.

01ESP32-S3
02Transformer
03C/C++
04PSRAM
05TinyML
05

IMPLEMENTATION

How it works in practice.

  1. 01

    Natural-language command -> tokenizer/model in PSRAM -> structured command -> GPIO/PWM/servo executor.

  2. 02

    Inference runs entirely on the ESP32-S3 with no cloud API and no PC in the loop.

  3. 03

    The project targets one known hardware profile instead of pretending to be universally portable.

06

TESTING & VALIDATION

How I know it works.

TEST 01

Held-out validation for English and Spanish command sets

TEST 02

Exact-match validation reported by the project

TEST 03

On-device inference on ESP32-S3-WROOM-1-N16R8

TEST 04

Real GPIO/PWM/servo action execution

TEST 05

Negative boundary documented for unsupported boards/domains

07

EXTERNAL SERVICES & DEPENDENCIES

What sits outside the core.

COREESP32 Command LLM
08

ENGINEERING EVIDENCE

Artifacts behind the claims.

01Training pipeline
02English and Spanish models
03On-device inference firmware
04Validation results
05GPIO command examples
09

BOUNDARIES / CONSTRAINTS

What the project does not pretend to solve.

Closed-domain model, not a general-purpose LLM

Requires PSRAM

Pin map targets one known module profile

No cloud fallback by design

10

RESULT / CURRENT STATE

What this produced.

A working proof that a small closed-domain language model can provide useful natural-language control entirely on a low-cost microcontroller.