Resources

Community

Start with Leo

How Israel's Elite Engineering Units Cut Months of Work to Minutes with Leo AI During Wartime

IMPACT IN NUMBERS

90%+

Time saved across validated engineering tasks

Days → 2 min

Structural reinforcement of a fielded combat vehicle

2 months → 2 weeks

Multi-system engineering integration, end to end

~8,500 hrs

Build hours behind a single program, where one mistake is costly

INDUSTRY

Defense, Automotive, and Armored Vehicle Systems

FUNCTION

Mechanical Engineering, Combat Vehicle Design, Restoration and Maintenance

TEAM SIZE

Engineering corps within two of the nation's largest defense engineering organizations

ENVIRONMENT

Active wartime operations, including the war with Iran. Zero tolerance for error.

Background

This is the story of how two of Israel's most demanding defense engineering organizations adopted Leo AI as a daily engineering tool and put it to work in the middle of a war. The first is MASHA, the IDF's Center for Restoration and Maintenance, responsible for maintaining, repairing, converting, and upgrading the platforms that fight, and the unit that manufactures the Merkava tank on its own production line. The second is RAPAT, the tank development authority and engineering arm of the Merkava and Armored Fighting Vehicle Directorate, responsible for the engineering design of the nation's heaviest armored platforms: the Merkava tank family, the Namer heavy APC, and the Eitan wheeled fighting vehicle.

The team includes commissioned officers with advanced engineering degrees, working complex mechanical and structural problems on platforms that operate under fire, and that are sometimes repaired in the field and behind enemy lines. The adoption did not begin at the top. It began with a major who ran the engineering corps and refused to accept that civilian engineers had AI while IDF engineers did not. He convinced his commander, then the commander of MASHA, then took it up the chain. Within roughly a year, Leo went from unknown to standard practice.

Then the war with Iran turned a productivity tool into an operational necessity. Engineering workloads spiked overnight. The teams were simultaneously managing vehicle maintenance demands, urgent design tasks, and operational support, all under the kind of pressure where cutting corners stops being a choice and becomes an inevitability when time runs out. These engineers had seen what happens when time pressure forces a system into the field without full calculation and validation. Those failures carry consequences that go well beyond cost overruns.

As the deputy CEO of MASHA put it, the engineering corps hit a workload so far out of proportion that it became almost impossible to work. Leo was not a nice-to-have in this context. It became the tool that allowed critical engineering work to get done at all.

Why General-Purpose AI Was Never an Option

Before the technical wins, it is worth addressing why an off-the-shelf chatbot was never going to work here.

A general large language model understands language and images. It does not understand an engineering reality: the parts, the past decisions, the standards, the manufacturing constraints. So it produces confident, fluent answers that are not grounded in anything real. In a domain where a wrong reinforcement calculation ends up on a vehicle carrying soldiers, confident-but-ungrounded is the worst possible failure mode. As Leo's CEO put it to the room: imagine letting a generic model loose on the design of a Merkava tank. It does not work.

Leo is different by design. It is grounded in the engineering knowledge base, trained on over a million pages of engineering standards, books, and technical literature, and connected to the organization's own knowledge base: its parts, its design history, its specifications and standards. Every answer is traceable to a verified source. Leo is SOC 2 certified and GDPR compliant. No AI is trained on customer data, customer data remains fully secure, and customer IP is never shared. It sits as an intelligence layer on top of the systems the teams already run. It does not replace them.

That is what made everything below possible.

The Challenge

The deputy CEO of MASHA described the constraint in three numbers and one sentence. Roughly ten engineers and analysts per engineering team. Around 8,500 build hours behind a single program. And a bar that keeps rising, because tomorrow's platform has to outperform today's to give soldiers a qualitative edge. His conclusion: even one small mistake costs us dearly.

When a problem falls outside a single engineer's specialty, or a critical reference cannot be found fast enough, the options are limited: spend days searching manuals and standards, wait for a senior expert to free up, or escalate outside the organization entirely. In active wartime operations, none of those timelines are acceptable. And the consequence of not doing the analysis is not a delayed project. It is a system that enters the field under-validated.

Two effects followed, both dangerous. Development timelines stretched longer and longer. And the cost of mistakes grew larger and larger. In a wartime environment, as the deputy CEO put it plainly, that simply does not work. An answer you cannot trace to a trusted source is an answer you cannot use.

What Changed

USE CASE 1

Structural Reinforcement of a Fielded Combat Vehicle

Captain Tomer, Mechanical Engineer

Time saved: days reduced to roughly 2 minutes. Risk reduction: a fielded vehicle returned to safe operational status without waiting on an unavailable analysis timeline.

"A problem came in from the field on one of our Eitan platforms. A structural component had developed a crack, and it needed to be reinforced properly, with the engineering to back it up. This is not the kind of thing you guess at. You have to understand the load path, the stress concentration at the crack, the right reinforcement geometry, and you have to be able to defend the calculation, because that vehicle goes back out with people inside it.

Normally a problem like this takes me anywhere from a day to a week. You go find the right standard, the relevant prior work, the governing equations, and then you model it. Under the workload we were carrying during the war, I did not have a week. I am not sure I had a day. So I pushed the problem into Leo from my phone.

In about two minutes, not a day, not a week, it surfaced the relevant equations, pulled the relevant papers and references from our engineering knowledge base, worked the problem with me, and gave me a 3D model of the reinforced part. Every step was sourced, so I could see where the answer came from and check it. That fix has already gone into production and is on its way to the field. Two minutes, from my phone, on a problem that used to eat my whole week."

USE CASE 2

Field Production Fix Under Live Operational Pressure

Major Yevgeny, Engineering Officer

Time saved: a normally multi-day validation-and-production cycle compressed to same-day. Risk reduction: full manufacturing validation preserved under wartime tempo, not skipped.

"There was a production-side fix we had to get right, fast, while forces in the field were depending on it. The temptation in that situation is always the same. The clock is running, so you cut the validation short and hope. We have all seen what happens when a system moves to manufacturing without the time to verify the design properly. The failures that come out of that are not abstract.

With Leo, we did the opposite. We confirmed the governing engineering equations, validated the change for manufacturing, and locked in a full production timeline before sign-off, and we still moved faster than we ever could have by hand. I took full responsibility for that solution, and I could take it with confidence because the analysis behind it was sourced and traceable, not a guess. The point was never speed for its own sake. It was speed with the rigor intact. Leo did not let us skip the validation step. It let us do it in a fraction of the time."

USE CASE 3

Multi-System Engineering Integration

Major Yevgeny, Engineering Officer

Time saved: roughly 2 months reduced to about 2 weeks. A single linkage that previously required reporting across a million siloed data points, produced in about 30 minutes.

"This is the one that changed how I think about what is possible. We needed to establish a high-level engineering linkage on the Eitan, connecting the operational requirements to the underlying engineering equations and to production. The way that normally works, you are manually reporting across a million siloed items spread over disconnected systems. The information sits on managers' desks. Nothing moves. It is a two-month effort, and that is when things go well.

Leo produced the core linkage in about half an hour, tying together the engineering data, the queuing, and the production orders. The work that had taken us two months across siloed systems, we redid through Leo in a fraction of that time once the systems were connected. And that is just the beginning of what we can do with it."

USE CASE 4

3D Model and Design Solution from a Field Input

Captain Tomer, Mechanical Engineer

Outcome: a production-ready 3D model and design solution generated from a single field-captured input, in minutes, usable directly in the existing engineering environment.

"One of the things that surprised me most is how little I have to give Leo to get something real back.

I can start from a photograph of a part, or a field description of what we need, and Leo will take it through to a 3D model and a design solution I can actually use. For a unit that maintains and adapts platforms that have been in service for years, often with limited digital design records, that is enormous.

It means a single engineer in the field, with a phone, can get to a validated, modeled solution without running the full manual design cycle for every part.

That is the difference between being stuck and moving forward. And because every output is sourced and traceable, I can stand behind it."

USE CASE 5

Grounded Calculations and Standards, Source-Cited

Major Noam, Engineering Division

Outcome: trusted, source-cited technical outputs replacing slow manual search through standards and references.

"What matters most to me in an environment like this is that the calculations are right and I can prove where they came from. Leo pulls from verified engineering sources and shows you the citation. When a question lands that used to mean hours digging through manuals and standards, Leo returns the governing equations and the relevant references in seconds, with the source visible so I can confirm it. For a unit maintaining platforms that have been in service for years, that is the difference between trusting an answer and second-guessing it."

USE CASE 6

Adoption Across the Engineering Corps

Major Noam, Commanding Officer, Engineering Division

Outcome: from a single major's initiative to standard daily practice across the engineering environment within roughly a year.

"What started as one major's refusal to let IDF engineers fall behind is now something we use every single day. The tool earns its place one solved problem at a time. An engineer brings a problem they have been stuck on, gets a sourced answer in minutes, and tells the next engineer. That is how it spread, not from a mandate, but from results that hold up when you check them."

Leadership Perspective

Major Noam, Commanding Officer, Engineering Division

"I want to be honest about where I started. I came into this as a commander, not as someone who was going to sit and run the tool myself. What changed my view was hearing my engineers describe, with specific numbers, the problems Leo had solved. A structural reinforcement on a fielded vehicle done in two minutes instead of a week. A multi-system integration we had budgeted at two months, redone in a fraction of that. A production fix validated fully, under fire, without cutting the corners that get people hurt. We adopted this in roughly a year, from something almost unknown to something we use every single day. This is not a pilot. A manager who will not use AI is simply going to be left behind. The graveyards of AI are full of clever tools that went nowhere. What made the difference here was leadership and the courage to actually put it to use. AI is one thing. In the hands of these engineers, it does the work of an entire team."

Results at a Glance

Problem

Previous Approach

With Leo

Time Saved

Structural reinforcement, fielded vehicle

1 day to 1 week, manual analysis

~2 minutes, sourced, with 3D model

~95%

Field production fix under pressure

Multi-day validation, or corners cut

Same-day, full validation

Days to hours

Multi-system engineering integration

~2 months across siloed systems

~2 weeks; core link in ~30 min

~95%

3D model from a field input

Full manual design cycle per part

Minutes, directly usable

~95%

Standards, equations, references

Manual search across manuals

Instant, source-cited output

Hours to minutes

Engineering corps adoption

One major's initiative

Standard daily practice

~1 year to org-wide

The Partnership Model

What stood out beyond the technical performance was the commitment Leo brought to making these teams operational under genuinely difficult conditions.

Leo's CEO, speaking after a swing through engineering teams in San Francisco, Sweden, and London, framed it this way:

"Seeing Leo in the hands of mechanical engineers anywhere is moving. But seeing it in the hands of women and men engineers in uniform, in the middle of a war, is something else entirely. This was not a hand-off of licenses. We configured the deployment so engineers could work inside their own secure environment, handled the setup end to end, and stayed engaged through every adoption challenge while the unit operated under wartime conditions. Innovation does not grow in a vacuum. Change and transformation come down to leadership. We succeed in protecting the engineers and the field units because of leaders willing to put this to work."

What Comes Next

The expansion path is specific and outcome-driven. The strongest near-term trigger is onboarding: compressing the months of institutional knowledge transfer that a newly assigned engineer normally needs into a tool a new officer can use productively on day one.

The second trigger is consistent, documented return: time saved on calculations and analysis, reduction in the risk of under-validated systems entering the field, and quality improvements that show up directly in engineering output. In an environment where the cost of a missed calculation is not a rework cycle but a vehicle system in the field with people inside it, that return is not theoretical.

STOP LOSING ENGINEERING KNOWLEDGE

Every resignation email is a knowledge transfer you are not ready for.

Leo AI makes the institutional knowledge already in your CAD vault searchable by every engineer on your team, before the next senior engineer walks out the door.

Schedule a Demo →

#1 New AI Software Globally - G2 2026

Enterprise-grade security

Trusted by world-class engineering teams

Subscribe to our engineering newsletter

Be the first to know about Leo's newest capabilities and get practical tips to boost your engineering.

Need help? Join the Leo AI Community

Connect with other engineers, get answers from our team, and request features.