Meta Model API

CASE STUDY

Helping developers manage high-stakes, high-security codebases

At a glance

Industry

Software

Use case

Streamlining legacy codebase management and modernization

Goal

Assist developers with massive codebases and automate mundane coding tasks

Llama version

Llama 3.1 70B Instruct

Deployment

Sourcegraph’s code intelligence platform, AWS LISA

Download full case study

THEIR STORY

Making smart smarter

Leidos serves government and commercial customers with smarter, more efficient digital and mission innovations. Headquartered in Reston, Virginia, Leidos helps solve the toughest challenges in healthcare, national security, transportation and defense.

THEIR GOAL

Automating code discovery, management and generation for massive, legacy codebases

Leidos manages, modernizes and authors software for defense, national security and healthcare clients. They grapple with decades of legacy code and work in extremely secure, air-gapped environments.Modernization is key, but the risk to the critical services means the codebases and proposed changes need to be thoroughly documented, analyzed and understood. The team needed an agentic toolset that could answer questions about codebases, automate code modernization and documentation and help debug.

THEIR SOLUTION

Llama and Sourcegraph power comprehensive codebase insight and code generation

Leidos uses Sourcegraph to manage and modernize legacy codebases. The Sourcegraph platform turns codebases into knowledge bases, which the platform’s AI uses to help developers understand legacy software and generate accurate code.Leidos paired Sourcegraph with Llama 3.1 70B Instruct, creating a powerful solution that helps developers generate unit tests, create code skeletons and document code. The solution can answer complex questions about code details — a huge benefit for developers ramping up on unfamiliar projects.

THEIR APPROACH

Security and cost constraints required an open-source LLM, only Llama proved fully capable

Leidos developers work in highly secure environments running on Amazon Web Services (AWS) Logically Isolated Secure Air-gapped (LISA) infrastructure, which rules out API calls to hosted models. Using proprietary models locally was a no-go due to per-token charges. Open-source large language models (LLMs) seemed to be an ideal solution, but tests demonstrated two major flaws: high latency and out-of-date training data.“With other open-source models, we were experiencing significant lag. Our Sourcegraph clients would timeout,” says Jay Davis, Chief National Security Software Architect at Leidos. “One model had no knowledge of Java beyond v15. Once we implemented Llama, timeouts stopped and the model’s code libraries were up to date.”

THEIR SUCCESS

Open-source Llama helps Leidos innovate within tight budgets

With Llama and Sourcegraph, Leidos found an open-source technology they could use to bring leading generative AI innovation to customers with fixed budgets who don’t invest in anything that doesn’t advance the mission.Llama-powered innovation has proven to be a valuable asset. Their success in combining Llama and Sourcegraph has given Leidos a competitive edge in federal contracts for advanced software development.

+20% increase in developer efficiency

Please confirm your identity

Please confirm your identity

LLAMA 3.1 70B INSTRUCT

LEIDOS

+20%