Build hardware that
speaks your language.
The agentic framework bridging low-resource African languages directly to physical IoT microcontrollers.
What is ASHA?
Ongoing research focused on bridging native African dialects to physical computing.
ASHA is an agentic framework designed to democratize smart hardware control for speakers of low-resource languages, starting with Twi in Ghana.
Commercial smart assistants almost exclusively serve dominant global languages. ASHA solves this by coupling fine-tuned indigenous speech recognition with an agentic LLM layer that absorbs high transcription noise, converting everyday mother-tongue speech into direct, bus-level microcontroller actions.
Vernacular Speech
Transcribes native Twi dialect audio commands directly.
Agentic Reasoning
Absorbs ASR errors and validates real-time device states.
Edge Hardware
Drives relays, sensors, and servos on ESP32 microcontrollers.
Projects Built With ASHA
Explore real-world hardware, IoT automations, and edge AI applications powered by the ASHA architecture.
ColdGuard
Hackathon WinnerAn AI-powered cold chain guardian protecting life-saving medicines. Built on ASHA's edge hardware architecture to monitor temperature telemetry and trigger fail-safes both online and completely offline.
Cinderella
Gemma ChallengeAn agentic IoT system pairing Google Gemma with ASHA's Model Context Protocol (MCP) server and ASHA.h firmware. Empowers open models to visually perceive environments and orchestrate physical hardware autonomously.
Smart Home Testbed
Physical PrototypeA physical multi-zone scale testbed demonstrating multilingual voice control in Twi and English. Validates real-time contextual reasoning, automated perimeter security, and bus actuation across PWM, digital, and analog circuits.
Declare your devices with a single clean call.
No custom networking or complex cloud synchronization code needed. Simply declare your device category, name, and hardware bus directly in firmware.
#include <ASHA.h>
// Register hardware in one declarative line
asha.asha_devices.addDevice(
asha.genericDev(DeviceCategory::Actuator, "My Light", BusType::PWM),
18
);
// Initialize cloud agent bridge
void setup() { asha.begin(); }
Empirical Proof
Tested rigorously in the lab, proven across physical hardware testbeds, open models, and real African vernacular speech.
100% Task Success
Flawless end-to-end device actuation across hundreds of multilingual test runs and hardware pins.
Bring Your Own LLM
Cloud, hybrid, or 100% offline. Plug in open-weight models locally or connect to foundation APIs.
Sub-Second Dispatch
Eliminates slow cloud roundtrips. Instant execution from vernacular speech directly to edge pins.
Modular Languages
Add new regional dialects and languages in minutes without touching low-level microcontroller firmware.
Read the Full Research Paper
Official IEEE ICAST 2026 submission with full methodology, schematics, and testbed logs.
Early Access Opening Soon!
Be first in line to test the firmware, deploy WhatsApp voice agents, and connect low-resource IoT devices.
Zero spam. No proprietary lock-in. Open for developers & researchers.