About Naqi Logix
Naqi Logix builds a neural earbud and micro-gesture interface platform. It lets people move through the digital world using the quiet signals of the body instead of a tap of the finger or the sound of a voice. A clenched jaw, a small turn of the head, the faint electrical weather of muscle and brain: each becomes a signal, and each signal becomes intent.
Our work sits where physiology meets computation. The earbud stays in the ear for hours. It has to listen closely, decide quickly, and use very little power while doing it. We are moving from a single- signal world toward a multi-biosignal one, where muscle, motion, and neural activity are read together and understood as one continuous stream of intent. We are building the team that will carry this work from prototype to scale.
The Work
You will own the design, training, review, and compression of convolutional and related deep learning models for biosignal time series. That starts with EMG and IMU and, as the platform grows, extends to EEG and other physiological signals. You will work closely with firmware, hardware, and research colleagues so that what works in a notebook also works in the ear. As the team grows, you will help set its modelling standards and guide junior colleagues.
What You Will Do
- Design and iterate on architectures for detecting gestures, intent, and states from multichannel biosignal streams. These include 1D CNNs, temporal convolutional networks, and lightweight hybrid models.
- Own experiments from start to finish, from the first hypothesis through data preparation, training, evaluation, and documentation
- Review existing model and signal-processing code and lead improvements to it, including the detection pipelines for jaw-clench and head-gesture input.
- Design preprocessing pipelines: filtering, artifact handling, windowing, normalization, and feature extraction suited to each signal type.
- Build evaluation methods that fit the realities of biosignal data. These include differences between subjects, drift across sessions, differences in electrode contact and fit, and leakage between training and test sets.
- Compress and deploy models for edge inference within real limits on latency, memory, and power. You will use quantization, pruning, and efficient architecture choices, and you will work with firmware engineers on how models are integrated.
- Turn synchronized multi-device validation study data into training and benchmark datasets.
- Use Claude Code and other frontier coding tools to speed up exploration, testing, and refactoring. Review AI-generated code as carefully as you would review a colleague's pull request, and help define how the team does this.
- Document experiments so they can be reproduced and compared, and mentor junior team members as the team grows.
What You Bring
- 2–4 years of hands-on deep learning experience in industry or applied research.
- At least one model you have taken past the prototype stage into a product, a device, or a
- validated research result.
- Strong Python and deep fluency with PyTorch and/or TensorFlow/Keras.
- Solid experience with time-series or signal data, and working knowledge of digital signal processing (sampling, filtering, spectral analysis).
- A background in computer vision or another deep learning field, and the ability to apply it to sequential and sensor data.
- Experience fitting models to resource limits on mobile, embedded, or edge targets.
- Regular, practical use of AI coding assistants or agentic tools such as Claude Code in your work.
- Sound instincts about evaluation. You are suspicious of results that look too good, and you know how to check them.
- Strong habits with Git, code review, and collaborative engineering.
- Nice to Have
- Experience with EMG, EEG, ECG, IMU, or other wearable sensor data.
- Deployment to microcontrollers, using tools such as TensorFlow Lite for Microcontrollers,
- ONNX Runtime, ARM CMSIS-NN, or vendor toolchains.
- Quantization at INT8 and below, and benchmarking models on the device itself.
- Familiarity with Lab Streaming Layer (LSL), BLE data transport, or synchronizing multiple devices.
- Experience working with human-subjects research data and the care it requires.
- Publications, patents, or open-source contributions in signal processing or ML.
- A record of building and maintaining data or ML pipelines that others relied on.
Who We Hope You Are
- Collaborative by default. You share early, ask for review, and give review generously. You would rather the team be right than be right alone.
- Rigorous with fast tooling. You move quickly with AI assistance and slow down where it matters: at the evaluation, the data split, and the signal, artifact
- Comfortable with ambiguity. Our product and science are still taking shape. You can work in open territory without losing the thread
- Honest about uncertainty. You say what you know, what you suspect, and what still needs testing.
Why Naqi
- Work on a new category of human–computer interaction while it is still being formed.
- Help decide how a small, AI-native team builds products at scale.
- Work with research partners to validate real neural and physiological signals.
- Own meaningful problems from your first week.
