OpenBOM and Leo AI: Making Product Data Intelligent — Not Just Stored

Company Profiles

OpenBOM was founded by Oleg Shilovitsky with the explicit goal of providing BOM management and product data collaboration to hardware companies that are too small for enterprise PLM but too complex for spreadsheets. Shilovitsky brings a PLM industry background — he has observed from the inside how legacy PLM systems create friction rather than value for smaller teams — and OpenBOM reflects that accumulated frustration turned into product design. The platform is cloud-native, multi-tenant, and designed to be in productive use within days of signup.

Leo AI was founded by Maor Farid, who came from an engineering background before becoming an AI researcher. The company applies machine learning to a specific engineering problem: multi-objective design optimization. In hardware development, good decisions almost always require trading off competing objectives — lighter weight versus lower cost, higher performance versus simpler manufacturing, more reliable versus faster to produce. Leo AI's platform finds designs on the Pareto frontier of these trade-offs — the set of designs where you cannot improve one objective without making another worse — by evaluating thousands of design alternatives simultaneously.

Both companies share a core philosophy: AI in engineering should augment engineer judgment, not attempt to replace it. The engineer defines the objectives, the constraints, and the acceptable trade-off space. The AI explores that space at superhuman speed and presents options the engineer would not have had time to discover manually.

---

The Challenge

The SMB PLM Gap

Enterprise PLM — Teamcenter, Windchill, 3DEXPERIENCE — is designed for programs with 50+ engineers, complex BOM structures, multi-site manufacturing, and regulatory requirements that justify a multi-year implementation. The cost and complexity are appropriate for that scale.

For a hardware startup with 5–20 engineers, or a manufacturing company with 20–100 people, enterprise PLM is the wrong tool. It is too expensive, too complex to configure, and too slow to deploy. The typical result: BOM management in Excel, revision control via Dropbox folder naming conventions, and supplier communication via email.

This works until it doesn't. The failure modes are well-documented: wrong revision sent to a supplier, uncontrolled changes creating configuration management chaos, inability to know what shipped to customers if a product recall is needed. Every hardware company that grows past 10 engineers discovers these problems.

OpenBOM targets this gap with a philosophy that reflects what a small team actually needs: BOM management that works immediately, collaboration that doesn't require training, pricing that doesn't require a procurement process, and integration with the CAD tools (SolidWorks, Fusion 360, Onshape) that small teams already use.

Multi-Objective Optimization: The Human Limitation

Engineering design decisions are inherently multi-objective. A structural member needs to be strong enough (minimum performance), light enough (weight constraint), manufacturable (process constraint), and cheap enough (cost constraint). These objectives conflict. Making it stronger typically makes it heavier. Making it cheaper typically makes it harder to manufacture with tight tolerances.

Human engineers navigate this by experience and intuition — a good engineer develops a sense of the trade-off space in their domain over years of practice. But intuition is bounded: a human can hold a handful of design alternatives in mind simultaneously. The Pareto frontier of a multi-dimensional optimization problem may contain thousands of solutions, each representing a different balance of trade-offs.

Leo AI evaluates the full solution space — not the handful of alternatives an engineer could manually compare. The result: engineers see trade-offs they would not have discovered, and make design decisions with full information about the compromise they are accepting.

---

What OpenBOM Built

Live BOM Collaboration Without File Sharing

OpenBOM's core product is a cloud-native BOM database with real-time multi-user access. The fundamental difference from spreadsheet-based BOM management: there is one BOM, visible to everyone with access, always current.

This matters in specific ways for small hardware companies:

Contract manufacturing: When a hardware company works with a contract manufacturer (CM), both parties need to work from the same BOM. Email-based file sharing creates version confusion. OpenBOM allows the CM to access the live BOM directly, see changes in real time, and add manufacturing-specific data (lead times, supplier quotes) without creating a fork.

Revision control: OpenBOM's revision system is explicit — a part or assembly can only be released at one revision at a time, changes flow through a formal process, and the history is permanent. This is the minimum viable configuration management that prevents the "which version did we ship?" problem.

Supplier collaboration: Supplier quote requests, make/buy decisions, and approved vendor lists are managed within the BOM rather than in separate spreadsheets, keeping supply chain data connected to the product record.

CAD integration: OpenBOM integrates with SolidWorks, Fusion 360, Onshape, and other CAD tools to pull BOM data automatically from the design model, eliminating the manual data entry that creates errors in spreadsheet BOMs.

The deployment reality: most OpenBOM customers are in productive use within 1–3 days of signup. Enterprise PLM deployments for the same company would take 6–18 months.

---

What Leo AI Built

Leo AI's platform is designed for a specific workflow moment: when an engineer has a design problem with multiple competing objectives and needs to find a good design without exhaustively evaluating all possibilities.

The platform works in four stages:

Problem definition: The engineer defines the design space (the parameters that can vary and their ranges), the objectives (what to optimize — minimize weight, minimize cost, maximize strength, minimize manufacturing time), and the constraints (what must not be violated — maximum displacement under load, minimum factor of safety, maximum outer dimensions).

Design of experiments: Leo AI generates an initial set of design experiments that efficiently samples the design space, typically 50–200 initial evaluations using a space-filling design strategy.

Model building: Using the initial evaluation results, Leo AI trains a surrogate model — a fast-to-evaluate approximation of the relationship between design parameters and objective values. This model is accurate enough to guide optimization without requiring a full physics simulation for every candidate.

Pareto optimization: The surrogate model is used to explore the design space and identify the Pareto frontier — the set of designs where no further improvement in one objective is possible without degrading another. Engineers review the Pareto frontier and select the design that best represents their actual priorities.

The output is not a single "best" design. It is a range of good designs with explicit trade-off characterization. This is the right output for engineering decisions, where the "best" design depends on priorities that the tool cannot know — whether the program is weight-critical (aerospace) or cost-critical (consumer goods).

---

Results

OpenBOM outcomes:

• Time to productive BOM management: 1–3 days vs. 6–18 months for enterprise PLM
• Version control errors: customers report near-zero "wrong revision sent to CM" incidents after OpenBOM deployment, versus the regular occurrence before
• Supplier collaboration: quote request and approval workflows that previously took 2–3 email rounds are completed in the platform with a single notification
• Revision tracking: customers with product recall concerns can identify what configuration shipped to which customer from OpenBOM's history — a capability that did not exist with spreadsheet BOMs

Leo AI outcomes:

• Design space exploration: engineers using Leo AI report evaluating 10–50x more design candidates per development phase vs. manual comparison
• Time-to-concept: companies deploying Leo AI for new component design report 25–40% reduction in time from design brief to approved concept
• Risk reduction: by making trade-offs explicit, Leo AI reduces the frequency of late-stage design changes driven by discovering a constraint that was not visible during manual design exploration
• Cost optimization: for programs with explicit cost targets, Leo AI identifies cost-performance Pareto points that manual exploration would have missed, enabling designs that hit cost targets without sacrificing required performance

---

Lessons Learned

1. The SMB hardware market is larger than PLM vendors think. OpenBOM's success reflects a genuine market need — hardware companies below the enterprise threshold that need real PLM capabilities, not just better spreadsheets. The segment is large, fast-growing, and underserved.

2. Days vs. months is the real competitive differentiator for SMBs. Feature parity with enterprise PLM is not what OpenBOM sells. Speed to value — in productive use in days rather than months — is what small hardware companies actually buy.

3. Multi-objective optimization is the natural AI fit for engineering. Engineering decisions always involve trade-offs. An AI that helps navigate trade-offs is helping with the actual work of engineering. An AI that finds a single "optimal" answer is oversimplifying a problem that doesn't have a unique answer.

4. Pareto frontiers are more honest than single-point recommendations. Showing engineers the trade-off space and letting them choose based on their actual priorities is both more accurate and more trustworthy than an AI that picks a design for them.

5. Cloud-native and API-first are not features — they are architectural requirements for the modern hardware stack. Small hardware companies use SaaS tools for everything. A PLM system that requires manual data entry, can't integrate with their CAD tool, and doesn't have an API is architecturally incompatible with how they work.

---

Implementation Advice

For hardware startups and small manufacturers:

If you are managing BOMs in Excel, you are one revision error or one supply chain crisis away from a painful problem. OpenBOM's payback period is measured in the first incident it prevents, not in productivity improvement. Deploy it before you need it.

If you have design optimization problems — weight, cost, performance, or manufacturing trade-offs — Leo AI's entry point is a specific design challenge with at least two competing objectives and a willingness to define the design space formally. The output is most valuable when engineering leadership can articulate the trade-off priorities they are actually making.

Both tools work best in combination with other modern engineering tools (cloud CAD, collaborative project management, cloud PLM for configurations) rather than as islands in a legacy tool stack.

---

About the Source

This case study is drawn from AI Across the Product Lifecycle Episode 5, a podcast conversation with Oleg Shilovitsky (OpenBOM) and Maor Farid (Leo AI). See also: [[OpenBOM Review]], [[Cloud PLM vs Enterprise PLM]], [[PLM for Hardware Startups]], [[PLM Glossary: BOM]].