Some things I’ve worked on
This is a non-comprehensive list of some things I’ve done and am proud of.
re-mind
Personal project
An agentic chatbot that helps resurface ideas, goals, and intentions to encourage meaningful follow-through, guided by cognitive and behavioral research.
More info
re-mind is a way to take back control of what pops up in your mind.
It determines when, how and how often to resurface ideas, goals, and intentions to encourage you to act on them.
Developing with Nicole. Building with PydanticAI, FastAPI, Celery, Redis, and PostgreSQL, deployed as a chatbot for Telegram/WhatsApp. Currently in closed testing, check back soon for early access :)
Tire Slip Detection
At Mobileye
A patent-pending system for high-accuracy detection of lateral tire slip from images and vehicle signals. Provides a novel capability for detecting traction loss, previously unattainable with traditional sensors.
More info
Led end-to-end R&D effort, alongside junior engineer and senior researchers across the division. Approach fuses vision cues with vehicle dynamics to detect lateral traction loss in real time. Validated for real-world use across multiple vehicle types and environments, using diverse benchmarking strategies. Achieved 2x better accuracy than projected. Secured strategic deal commitments from VW and Porsche for integration into next-generation safety platforms.
Road Profile
At Mobileye
Premium perception product for high-precision estimation of road surface from image data. Enables dynamic suspension systems for improved comfort and safety in top-tier ADAS programs.
More info
Took long-term ownership and modernized this massive legacy ADAS system after more than a decade in production. Collaborated closely with OEM engineers to discern market requirements and deliver new features, resolve issues, and align future development. Introduced first empirical evaluation framework using LiDAR-based ground truth to quantify real-world accuracy. Redesigned the system around an event-based architecture, aligning with market preference and significantly reducing support overhead. Successfully converted customers to the new system.
ParallaxNet
At Mobileye
Patent-pending deep neural network for image-based 3D reconstruction in autonomous driving. Deployed in millions of vehicles across multiple OEMs like Porsche, Hyundai, and Kia.
More info
Led the development across all stages of the project, alongside a junior engineer, replacing an unmaintainable legacy system. Guided research direction, evaluation methodology, infrastructure design and engineering. Trained using Tensorflow/AWS on LiDAR-derived GT, with inference rewritten and optimized in C++ for on-chip deployment. Delivered a robust model with 52% higher accuracy and 2× faster runtime. This became the first in-production ML model developed fully in-house within traditionalist Geometry algo group, paving the way for broader adoption across the org.
Construction Optimization Modeling
Stealth startup
Real-time optimization pipeline for an early-stage startup, minimizing project cost and turnaround time in large-scale construction.
More info
Conducted applied research, designing and implementing a multi-stage optimization framework for resource allocation and execution efficiency. Integrated diverse optimization techniques with real-world engineering constraints to achieve near real-time performance on large-scale problem instances.
Delivered a robust production system that became a core part of the company’s offering, enabling early customer traction ahead of MVP - driving measurable cost and time savings.
Beacon
Student hackathon entry
Offline mesh-networking app for disaster scenarios that relays peer-to-peer distress messages to reach search and rescue teams without internet connectivity. 3rd place winner at the 2019 HUJI Hackathon.
More info
Led the team through ideation, system design, architecture, and implementation of an Android app prototype. Built a fully functional demo in 24 hours, handling device-to-device message relay over BLE (via Hype SDK, now Uplink) - able to reach devices across multiple rooms with no wifi or cellular signal.