Archive position — measured, not model output
2 likes on Devpost
221 of the 7,856 archived projects have more likes, and 285 share exactly 2 — so this project's #312 place in the like-ranked listing is a tie-break inside that group, not a ranking.
Projects (log scale)
Likes on Devpost. ▲ marks this project's group.
Show the figures
| Likes | Projects | Share of archive |
|---|---|---|
| 0 | 5,592 | 71.2% |
| 1 | 1,758 | 22.4% |
| 2 | 285 | 3.6% |
| 3–4 | 132 | 1.7% |
| 5–9 | 75 | 1.0% |
| 10+ | 14 | 0.2% |
Executive Summary
What the company appears to be
EM-assist is a self-reported desktop electromagnetic (EM) simulation tool that integrates an AI agent (GPT-5.6) with a verified finite-element method (FEM) solver. The system allows users to describe EM design goals, such as "a cavity resonant at 8.2 GHz", and have the AI propose parameters grounded in physics, which are then validated by the numerical solver.
What changed
The project evolved from an idea sparked by a failed simulation during a research internship into a desktop application that combines a verified FEM core with an AI design agent. A key change was replacing a hardcoded heuristic with a real GPT-5.6 API integration, after discovering the initial version did not use the AI as intended.
Single most important open question
Is the AI agent actually driving the solver in a way that produces novel, validated designs, or is it merely a UI layer that reports pre-computed results?
What The Product Actually Is
The description states that EM-assist Studio is a cross-platform desktop electromagnetic simulator built on a finite-element core using Nédélec edge elements. It includes:
- Lumped ports and RLC boundary elements
- Full S-parameter extraction with Touchstone export
- Analytical single-mode waveguide ports
- Arbitrary-geometry import through Gmsh's OpenCASCADE kernel
- A small ONNX surrogate for fast pixelated-patch design search
The AI component, GPT-5.6, is described as a "real design agent" that proposes RF/antenna parameters and drives an actual numerical optimizer against the solver until convergence. It does not just report numbers but only reports results computed by the solver.
Evidence
- The author states: “EM-assist Studio is a cross-platform desktop electromagnetic simulator built on a finite-element core…”
- The author states: “GPT-5.6 sits on top as a real design agent: describe a target — 'a cavity resonant at 8.2 GHz' — and it reasons about the physics, proposes parameters grounded in real design equations, and drives an actual numerical optimizer against the solver until the result converges.”
- The author states: “It never just tells you a number. It only ever reports one the solver has actually computed.”
Inference The system is described as being built with Codex, with a deliberate phased approach to development, where each feature is tested before integration.
Positioning & Claim Evolution
The project is positioned as a desktop EM simulator that bridges AI-generated design ideas with verified physics. It is described as an alternative to commercial tools that are locked behind expensive licenses.
Evidence
- The author states: “AI can clearly find real electromagnetic designs humans wouldn't try, but the tools that could combine that with a solver you can actually trust are locked behind five-figure licenses.”
- The tagline states: “A desktop EM simulator where GPT-5.6 is a real design agent — it proposes RF/antenna parameters, a verified FEM solver checks every one.”
Inference The positioning evolved from an idea rooted in personal experience with simulation failures to a tool that aims to make AI-assisted EM design accessible and trustworthy.
Target Customer & ICP
Not evidenced. The description does not specify the target customer or ideal customer profile (ICP). It does not name industries, roles, or use cases beyond general EM design.
Evidence
- No mention of specific user personas, industry verticals, or job functions.
- The author’s personal experience is described but not generalized to a segment.
Business Model & Pricing Evidence
Not evidenced. There is no information in the description about pricing, monetization strategy, or business model.
Evidence
- No mention of revenue streams, pricing tiers, or commercialization plans.
- The project is described as a hackathon submission with no indication of a go-to-market strategy.
Technical & Delivery Signals
The system is built using:
- C++, CMake, Python, Rust, TypeScript, React, TailwindCSS, Tauri
- GPT-5.6 (via API)
- ONNX for surrogate modeling
- Gmsh and OpenCASCADE for geometry import
- Codex for development
- A verified FEM solver with convergence testing against manufactured solutions
The author emphasizes that:
- Every change to the solver ships with a test that would fail if incorrect.
- Nothing is marked "supported" unless it has been verified.
- The AI agent was replaced with a real GPT-5.6 API integration after discovery of a hardcoded heuristic.
Evidence
- The author states: “The entire project was built with Codex, in deliberate phases…”
- The author states: “Any change to solver physics ships with a test that would fail if the change were wrong.”
- The author states: “Catch the fake agent before it shipped, and replacing it with genuine GPT-5.6 tool-calling…”
Inference The technical stack suggests a desktop application with strong engineering rigor, but no evidence of delivery to end users or production deployment.
Traction & Maturity Signals
Not evidenced. There is no mention of customers, usage data, revenue, or product adoption.
Evidence
- No mention of user base, customer acquisition, or product performance metrics.
- The project is described as a hackathon submission and not yet deployed for public use.
Competitive Context
Not evidenced. The description does not compare EM-assist to existing tools or competitors in the EM simulation space.
Evidence
- No mention of competitors, market positioning, or differentiation from other tools.
- The author only references commercial tools as being expensive but does not name them.
Key Risks & Red Flags
- AI Integration Uncertainty: The system replaced a hardcoded heuristic with a real GPT-5.6 API integration after discovery — this suggests early-stage uncertainty in AI implementation.
- Unproven AI Value: The author states that an LLM is a poor EM solver but a good orchestrator, implying the value of the AI layer is not yet fully demonstrated.
- Limited Scope: The system currently lacks support for open-radiation boundaries and multi-mode waveguide ports — features critical to antenna design.
- Self-Reported Only: All claims are self-reported and unverified; no third-party validation or traction data exists.
Evidence
- The author states: “The most honest one: partway through, I found the 'AI design assistant' wasn't calling GPT-5.6 at all — it was a hardcoded heuristic…”
- The author states: “The moment we stopped asking GPT-5.6 to guess at physics and started asking it to choose objectives and parameters for a real deterministic optimizer to check, the system got more capable and more trustworthy.”
Diligence Questions To Ask The Founders
- What is the actual performance of the AI agent in generating valid designs? Can you show examples where it proposed something novel that was validated by the solver?
- How does the system handle real-world design constraints, such as manufacturing tolerances or material properties?
- Are there any plans to monetize or commercialize this tool? If so, what is the business model?
- What are the limitations of the current FEM solver and how do they impact usability for real-world applications?
- How is the ONNX surrogate used in practice — is it a full replacement or a fast approximation?
Investment/Partnership Verdict
Not evidenced. No information is provided about funding, valuation, or investment interest.
Evidence
- No mention of funding rounds, investors, or partnership opportunities.
- The project is described as a hackathon submission with no indication of commercial viability or investor interest.
Source
Submitted to the OpenAI 2026 hackathon on Devpost. Project home on DevPost.
The analysis above was generated by a language model from the project's own one-line description. It is not independent research and contains no verified traction, revenue or customer data.