Armic (@projectarmic) header
Armic (@projectarmic) profile picture

AI agents that manage patient recovery autonomously Onchain rehab + robotics

Tokyo-to, Japan
Joined April 2026
Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)Photo by Armic (@projectarmic)
Rehabilitation is broken. Patients drop out. Progress isn’t measurable. Outcomes aren’t trusted. We’re fixing this with AI agents.
18
3
2
57
16,154
ARMIC weekly build update 🛠️ This week we’re connecting exercise detection to the agent layer. Detecting movement is useful. But the real step forward is turning those detections into session context the AI can reason about. In this demo: → edge device detects rehab movements → recent exercise activity is passed to the agent → agent reviews the last movements → user receives feedback and next-step guidance This is how ARMIC starts moving from raw sensor data to adaptive rehabilitation workflows. Detect → Understand → Guide 🤖
8
5
1
21
1,660
Housekeeping is also underway. Better docs. Cleaner metadata. More transparency. Stronger public foundations. ARMIC is still early, but we want the project easier to verify as it grows.
1
3
9
672
ARMIC weekly update 🛠️ This week we’re moving exercise recognition closer to the device. Instead of treating every AI task the same, ARMIC is starting to separate the stack: → lightweight edge models for fast movement detection → agents for context, reasoning, and adaptation The current model detects: → lateral raise → bicep curl → elbow flexion This matters because rehab systems need fast local recognition before they can support adaptive interaction. Edge models handle perception. Agents handle reasoning. That’s the loop we’re building 🤖
6
6
2
22
2,796
ARMIC can now detect movement at the edge. Next step: make the system respond to it. 🤖
1
1
10
1,018
ARMIC foundation update 🛠️ We’re moving from motion capture to exercise recognition. Using the wireless motion devices from our last update, we collected data for basic rehabilitation movements: → Bicep curl → Lateral raise → Elbow flexion Then we trained an Edge Impulse model to classify the exercises using IMU data. This is an important step for ARMIC: Raw movement data becomes structured context that future agents can understand. Perception → Motion Capture → Exercise Detection → Adaptive Rehabilitation
6
7
4
36
4,458
This is still part of the Foundation phase. The goal is to build ARMIC layer by layer: 🧠 Brain 👀 Perception 🧩 Exercise recognition 💪 Action More soon. Perhaps very soon.
2
2
17
1,068
ARMIC is learning to understand movement. Today we’re testing the first stage of our motion capture pipeline using wireless IMU devices and Edge Impulse. The goal: turn raw acceleration + gyroscope data into rehabilitation exercise detection. Before ARMIC can adapt to a patient, it needs to recognize how the patient is moving. Perception → Motion Capture → Adaptation 🤖
9
8
2
37
3,927
Foundation phase update 🛠️ Last week we demonstrated local Edge AI on the Arduino UNO Q. This week we're focusing on perception. We're experimenting with ECG signals, motion sensing, and MQTT-based data pipelines to help ARMIC better understand patient activity during rehabilitation sessions. The goal isn't just automation. It's building systems that can observe, understand, and eventually adapt in real time.
7
9
3
40
5,192
Foundation phase update 🛠️ We’ve started validating the Edge AI architecture behind ARMIC using the Arduino UNO Q + local SmolLM2 inference. Current setup: → llama.cpp running locally → MCU + MPU synchronization → Local WebUI → Real-time peripheral coordination → On-device AI execution This is one of the first steps toward an autonomous rehabilitation system capable of real-time robotic interaction directly at the edge. Walkthrough below 🤖
6
6
2
22
3,255
One of the main reasons we migrated from the ESP32 to the Arduino UNO Q was the need to keep both AI inference and hardware coordination on the same edge device. Instead of relying on cloud execution, the goal is to enable: • local inference • real-time motor control • synchronized MCU + MPU operation • adaptive robotic interaction This shifts ARMIC from rigid automation toward autonomous rehabilitation systems driven directly at the edge.
1
3
11
1,153
ARMIC Roadmap (Next 3 Months) 🛠️🤖 Phase 1 — Foundation • Expand hardware stack (Arduino + robotics) • Improve agent execution flows • Refine onchain payment logic • Continue open development updates Phase 2 — Physical AI • Integrate robotic arm interactions • Connect autonomous actions with onchain triggers • Test machine-to-machine coordination flows • Begin real-world automation demos Phase 3 — Deployment Layer • Expand Solana infrastructure • Improve interoperability • Launch more public demos • Move toward continuous real-world operation We’re building toward systems where AI agents, robotics, and programmable finance converge.
22
8
2
37
10,285
The project has grown, the market has spoken so we have to deliver now. - Will be updating more than Bi-weekly. - Documenting everything - All open-source In addition to this Blue checkmark ✅️
3
2
1
18
2,232
From onchain → to real hardware. Used part of the project fees to start expanding the physical AI side of ARMIC. Arduino setup just arrived. Robotic arm next. We’re interested in building systems that don’t just exist on charts, but interact with the real world 🤖
15
10
2
42
7,557
Arduino Q for the uninitiated. Pretty much a microcontroller with AI capabilities. Ordered quite a few.
5
3
2
19
2,872
We didn’t win. We’re still building. The fees from $ARMIC are going into hardware, platform development, and open-sourcing the system. We’ll also enable DEX liquidity so it becomes a community token. 3-month roadmap. Biweekly updates. This doesn’t stop here.
20
3
47
5,158
Congrats again to @flypraxis, awesome project!
7
17
2,847
8
5
47
3,862
Armic retweeted
Thesis on Armic (@projectarmic): @projectarmic is a legitimate early-stage DeSci project building autonomous AI agents for patient recovery — AI that tracks, adapts, verifies, and rewards rehab progress, paired with robotic execution and on-chain incentives. The team is fully doxxed: biomedical engineers Luis Eduardo Arevalo Oliver (@EddOliver_), Alejandro Sanchez Gutierrez, and Victor Alonso Altamirano Izquierdo (@outlay_pay), with real experience in healthcare hardware, IoT, AI agents, and Web3 systems (prior hackathon work and roles at Blank!t). Launched transparently on Solana via @EasyA_Kickstart (in collaboration with @dom_kwok and @kwok_phil), with immediate supply locking via @Streamflow_Fi, fee allocation for hardware/open-sourcing, and a focus on real builds over hype. In a $30B+ rehab market crippled by poor adherence, this combination of AI autonomy, robotics, and tokenized incentives has genuine product-market fit potential. High execution risk (still pre-product), but the doxxed builder team, on-chain transparency, and narrative alignment (AI agents + health + DeSci) make it one of the more credible micro-funding experiments in the space right now.
10
10
5
38
10,285