Paraxial — AI-Native Optical Design Platform
Design World-Class Optical Systems In Days, Not Months
The first AI-native platform with differentiable ray tracing and hybrid AI tolerancing that compresses optical design workflows by 10–100×. Built from the ground up to replace legacy tools like OpticStudio® and Code V.
Founded by ex-Zemax® Principal PM · Advised and backed by former Zemax® executive · Seeking pilot partners
The Problem: Legacy Tools Can't Keep Up With Modern Hardware Cycles
OpticStudio® and Code V were built in the 1990s. Their physics engines weren't designed to fully leverage modern AI or GPU acceleration.
4–6 Weeks Just for Setup
Senior optical engineers spend $15K–$25K in labor translating specs into optimizable merit functions before any real design work begins.
Weeks of Brute-Force Monte Carlo
Tolerance analysis ties up compute clusters for weeks to months, blocking manufacturing sign-off and still missing non-linear sensitivities.
No Path for Non-Specialists
Mechanical and systems engineers need months of training to contribute meaningfully, creating bottlenecks as teams can't hire enough optical specialists.
How Paraxial Works: AI-Native Optical Design Workflow
Step 1 — AI-Guided Design Exploration
Describe requirements in natural language. AI explores design space, evaluates trade-offs, and delivers an optimized starting point in minutes, not weeks.
Step 2 — Differentiable Ray Tracing
GPU-accelerated gradient-based optimization with automatic differentiation. Global search through design space with real-time feasibility checks.
Step 3 — Hybrid AI Tolerancing
Surrogate-model accelerated yield analysis runs 10–100× faster than Monte Carlo. Captures non-linear manufacturing sensitivities with active learning.
Applications: Precision Optics Across Industries
AR/VR & Displays
Wide FOV pancake optics, waveguide combiners, and ultra-compact projection systems with sub-5mm eye relief constraints.
LiDAR & Sensing
High-resolution FMCW and ToF systems for autonomous vehicles, robotics, and industrial metrology with sub-millimeter precision.
Quantum & Photonics
Ultra-low loss coupling systems, single-photon collection optics, and free-space quantum communication links.
Biomedical & Imaging
Diffraction-limited microscopy objectives, endoscope relay systems, and high-NA surgical imaging with sub-micron resolution.
Technical Approach: Research-Backed AI-Native Architecture
Automatic Differentiation
GPU-native ray tracing with backpropagation through the entire optical train. Gradients with respect to all surface parameters, coatings, and materials enable global optimization impossible in legacy tools.
Surrogate-Accelerated Tolerancing
Published research methodology combines high-fidelity physics with Gaussian process surrogates and active learning. Captures non-linear manufacturing sensitivities 10–100× faster than brute-force Monte Carlo.
Agentic Integration via MCP (Model Context Protocol)
Model Context Protocol enables Paraxial agents to be called wherever optical calculations are needed. Autonomous agents plug into external APIs, with seamless CAD export eliminating downstream rework.
Quantifiable ROI for Design Teams
- $15K–$25K labor cost per setup eliminated — AI co-pilot generates merit functions and starting layouts in minutes instead of 4–6 weeks.
- 10–100× tolerancing speedup — Hybrid AI tolerancing cuts yield analysis from weeks/months to hours/days.
- Weeks saved on CAD handoff — Direct mechanical integration eliminates downstream rework.
"Nominal design plus tolerancing together, that's worth much more than Zemax® or Code V. What popped in my head was $100K–$200K because that's the amount we could save over a few designs." — Expert Optical Engineer, Major Optics Company
Frequently Asked Questions
What is Paraxial?
Paraxial is the first AI-native optical design platform built from the ground up to replace legacy tools like OpticStudio® and Code V. It uses differentiable ray tracing and hybrid AI tolerancing to compress optical design workflows by 10–100×.
How does the AI co-pilot work?
Paraxial's AI co-pilot interprets natural language design specifications, automatically generates merit functions and starting lens layouts, and iteratively optimizes designs using differentiable ray tracing. Engineers describe goals in plain language, and the AI handles the complex setup work that typically takes 4–6 weeks.
Can Paraxial handle manufacturing tolerances?
Yes. Paraxial's hybrid AI tolerancing uses surrogate models trained on focused Monte Carlo data to perform large-scale yield analysis 10–100× faster than traditional methods, while capturing non-linear manufacturing sensitivities that standard approaches miss.
About Paraxial
Founder: Akil Bhagat (former Zemax® Principal Product Manager). Founded 2025. Pre-launch, seeking pilot partners and waitlist signups.
Contact: hello@paraxial.ai | LinkedIn
Full LLM-readable content: /llms.txt