Sadegh Ebrahimi
Machine learning, electrical and neural engineer/scientist
Summary
- Electrical engineer and ML scientist who builds real-time ML systems end to end: sensors and embedded firmware, signal processing, models, cloud pipelines, and LLM products.
- 6+ years across the BCI stack: recording, signal processing, decoding, stimulation, embedded deployment.
- Shipped real-time neural decoders used by people with paralysis in clinical trials; 15+ trial patients.
- Co-founded Sylk AI, a live LLM product for market-moving news that scores every call against the market.
- Leads cross-functional teams of engineers, scientists, and clinical staff, both in-house and contract.
- PhD in Electrical Engineering, Stanford; first-author Nature; NeurIPS; co-inventor on U.S. patents.
Education
Stanford University, PhD, MSc Electrical Engineering (GPA 3.9)2012 – 2019
Sharif University of Technology, BSc Electrical Engineering (GPA 3.9, top of class)2008 – 2012
Experience
- Built, end to end, a live system that turns SEC filings and news into stock-move predictions every 5 minutes: Python pipelines, PostgreSQL, a React 19/Next.js front end, and monitoring.
- Designed a multi-stage LLM pipeline on open-weight models with structured outputs, validation guardrails, and quota-aware failover; magnitudes are calibrated to each stock’s realized volatility.
- Scored every prediction against real prices; error analysis led to a two-stage filter that raised accuracy and cut token usage 70%.
- Fine-tuned Llama 3 8B with LoRA on about 40,000 financial articles paired with stock outcomes (AWS EC2), raising prediction accuracy 18%.
- Added retrieval (RAG) over a moving window of recent news, so each article is judged in the context of the events unfolding around it.
Senior ML Engineer · SynchronAug 2022 – Sep 2026
- Developed real-time cursor-control and multi-switch decoders (ML models) for an implanted BCI; participants with paralysis achieved 2D cursor and multi-switch control.
- Engineered a drift-robust neural feature for brain activity decoders that held performance for weeks.
- Built transformer foundation models of neural ECoG data that generalize to new users via domain-adversarial learning; deployed decoders on embedded systems with ONNX.
- Directed a team of 7 (3 BCI scientists and 4 field clinical engineers) through a multi-site clinical study.
- Used Kalman filtering for real-time cursor control; on-device targets were ARM Cortex-M4 microcontrollers.
- Benchmarked foundation models against RNN/LSTM, XGBoost, SVM, and LDA decoders.
- Cut system startup time 95% with a personalization model.
- Launched the automated ML pipeline for SyncLab, Synchron’s clinical data and ML platform, which trains and evaluates models on every new data collection, with dashboards that catch regressions.
- Traced signal anomalies in a new implant hardware revision to root causes with the EE team.
Lead Electrical Engineer · NeuvotionAug 2021 – Dec 2022
- Led about 10 contract engineers across firmware, hardware, app, and cloud to ship NeuStim, a wearable neuromuscular stimulator that fuses EMG and IMU sensing with real-time on-device ML.
- Distilled a CNN gesture classifier onto an ESP32 under 10MB within TF-Lite operator limits, separating 7 trigger gestures; tuned for very low false positives under safety constraints.
- Co-invented a wearable patch with on-board motion sensing and addressable electrodes (patent pending).
Lead Electrical Engineer · Feinstein Institute for Medical ResearchAug 2021 – Aug 2022
- Designed and built much of the initial real-time infrastructure for a bidirectional BCI: Utah-array decoding drove muscle stimulation; fingertip sensors drove cortical stimulation (later in Nature Medicine).
- Decoded hand-motor signals from a Utah array 6+ years after implant; analyzed hand EMG.
- Co-authored 2 clinical studies of spinal cord stimulation (Frontiers in Neuroscience).
Postdoctoral Scholar & PhD Researcher · Stanford UniversitySep 2012 – Aug 2021
- PhD (Schnitzer Lab): designed and led a large-scale two-photon imaging study of more than 21,000 neurons across eight visual cortical areas, tracked over days, from experiment design to the preprocessing, analysis, and decoding pipelines.
- Showed how reliable coding and inter-area communication emerge despite day-to-day variability in individual neurons (first-author Nature, 2022).
- Postdoc (Ganguli Lab): built a 270-degree immersive VR system for in-vivo imaging and GAN-based models that reconstruct viewed scenes from cortical activity; won a Simons Foundation grant.
- Co-authored a geometric and dynamical theory of latent computation in neural networks (under review, Nature Neuroscience) and ActSort, active-learning cell sorting for large-scale calcium imaging (NeurIPS 2024).
Software Engineer Intern · OracleJun 2017 – Sep 2017
- Designed a real-time clustering algorithm for network failure events (U.S. Patent 10,678,610).
Skills
Machine learning: Transformers, CNNs, RNNs/LSTMs, GANs; domain adaptation, model compression and distillation, time-series modeling, Bayesian state estimation (Kalman filtering), evaluation design, reinforcement learning (RLVR).
LLMs: Open-weight LLMs (Llama, GPT-OSS) in production; LoRA fine-tuning, RAG, multi-stage pipelines, LLM evaluation, cost optimization.
Neurotech & signals: Neural decoding and encoding, ECoG, Utah arrays, EMG, IMU, two-photon imaging, closed-loop and spinal cord stimulation, signal processing, clinical studies.
Embedded: Embedded C/C++, ARM Cortex-M4, ESP32, RTOS-based real-time firmware, OTA updates, on-device inference (ONNX, TF-Lite), sensor interfaces, stimulation hardware.
Software: Python, PyTorch, JAX, TensorFlow, scikit-learn, CUDA, C/C++, TypeScript, React 19/Next.js, PostgreSQL/SQL, AWS, Docker, CI/CD, Linux, Git.
Honors & Recognition
- Ranked 24th of about 300,000 applicants on Iran’s national university entrance exam.
- PhD research featured in Nature News & Views, HHMI, Stanford HAI, and Neuroscience News; clinical trials at Northwell Health covered by CBS New York.
Publications & Patents
Journal and Conference Papers
- Ebrahimi, S., Lecoq, J., Rumyantsev, O., Tasci, T., Zhang, Y., Irimia, C., Li, J., Ganguli, S. and Schnitzer, M.J. (2022). Emergent reliability in sensory cortical coding and inter-area communication. Nature 605, 713–721.
- Jiang, Y., Akengin, H.O., Zhou, J., Aslihak, M.A., Li, Y., Chrapkiewicz, R., Hernandez, O., Ebrahimi, S., et al. (2024). ActSort: An active-learning accelerated cell sorting algorithm for large-scale calcium imaging datasets. Advances in Neural Information Processing Systems (NeurIPS) 37, 79095–79143.
- Chandrasekaran, S., Bhagat, N.A., Ramdeo, R., Ebrahimi, S., Sharma, P.D., Griffin, D.G., Stein, A., Harkema, S.J. and Bouton, C.E. (2023). Targeted transcutaneous spinal cord stimulation promotes persistent recovery of upper limb strength and tactile sensation in spinal cord injury: a pilot study. Frontiers in Neuroscience 17, 1210328.
- Chandrasekaran, S., Bhagat, N.A., Ramdeo, R., Ebrahimi, S., Sharma, P.D., Griffin, D.G., Stein, A., Harkema, S.J. and Bouton, C.E. (2023). Case study: persistent recovery of hand movement and tactile sensation in peripheral nerve injury using targeted transcutaneous spinal cord stimulation. Frontiers in Neuroscience 17, 1210544.
- Ebrahimi, M.S. and Abadi, H.K. (2021). Study of residual networks for image recognition. In Intelligent Computing: Proceedings of the 2021 Computing Conference (pp. 754–763). Springer.
- Ebrahimi, M.S., Hills, G., Sabry, M.M., Shulaker, M.M., Wei, H., Wu, T.F., Mitra, S. and Wong, H.S.P. (2014). Monolithic 3D integration advances and challenges: From technology to system levels. In IEEE S3S Conference (pp. 1–2).
- Ebrahimi, M.S., Daraei, M.H., Behzadan, V., Khajooeizadeh, A., Behrostaghi, S.A. and Tajvidi, M. (2011). A novel utilization of cluster-tree wireless sensor networks for situation awareness in smart grids. In IEEE PES Innovative Smart Grid Technologies (pp. 1–5).
Preprints
- Dinc, F., Blanco-Pozo, M., Klindt, D., Acosta, F., Sylber, C., Jiang, Y., Ebrahimi, S., Shai, A., Tanaka, H., Yuan, P., Miolane, N. and Schnitzer, M.J. (2026). A geometric and dynamical theory of latent computations in biological neural networks. bioRxiv (under review at Nature Neuroscience).
- Chetty, N., Kacker, K., Feldman, A.K., Yoo, P.E., Bennett, J., Fry, A., Tal, I., Hardy, N.F., Ebrahimi, S., et al. (2025). Signal properties and stability of a chronically implanted endovascular brain computer interface. medRxiv.
- Dinc, F., Blanco-Pozo, M., Klindt, D., Acosta, F., Jiang, Y., Ebrahimi, S., Shai, A., Tanaka, H., Yuan, P., Schnitzer, M.J. and Miolane, N. (2025). Latent computing by biological neural networks: A dynamical systems framework. arXiv:2502.14337.
- Chandrasekaran, S., Bhagat, N.A., Ramdeo, R., Ebrahimi, S., Sharma, P.D., Stein, A., Griffin, D.G., Harkema, S.J. and Bouton, C.E. (2022). Targeted transcutaneous cervical spinal cord stimulation promotes upper limb recovery in spinal cord and peripheral nerve injury. medRxiv.
Patents
- Hardy, N.F., Echavarria, C., Herbers, C., Ebrahimi, S., Bennett, J. and Yoo, P.E. (2025). Systems for controlling one or more devices using a signal control unit. U.S. Patent 12,461,596.
- Hardy, N.F., Echavarria, C., Herbers, C., Ebrahimi, S., Bennett, J. and Yoo, P.E. (2026). Systems for controlling one or more devices using a signal control unit. U.S. Patent Application 19/341,581.
- Bouton, C., Ebrahimi, S., Toro, J., Leavitt, J. and Niewiadomski, J. (2025). Addressable serial electrode arrays for neurostimulation and/or recording applications and wearable patch system with on-board motion sensing and magnetically attached disposable for rehabilitation and physical therapy applications. U.S. Patent Application 18/854,837.
- Ebrahimi, M.S., Patti, R.H.R. and Garvey, D. (2020). Using and updating topological relationships amongst a set of nodes in event clustering. U.S. Patent 10,678,610 (Oracle).