Ertas for Code Generation

The Challenge

Software development teams are adopting AI-powered code completion, code review, and documentation generation at a rapid pace — but the productivity gains from generic code models plateau quickly on enterprise codebases. Internal frameworks, proprietary APIs, custom design patterns, and organization-specific coding conventions are invisible to models trained on public open-source repositories. The result is AI suggestions that are syntactically valid but architecturally wrong: importing deprecated internal packages, ignoring custom linting rules, or generating boilerplate that contradicts the team's established patterns.

Intellectual property is the deeper concern. Enterprise source code represents years of engineering investment and competitive advantage. Sending proprietary code to a third-party AI provider's API — where it may be logged, cached, or used in aggregate to improve the provider's models — creates IP exposure that legal and security teams are right to reject. Many organizations with strict IP policies have banned cloud-based code assistants entirely, forcing developers back to manual workflows while competitors accelerate with AI.

The Solution

Ertas lets engineering organizations build code AI that understands their codebase and runs entirely within their own infrastructure. Using Ertas Studio, platform engineering teams can fine-tune code-specialized foundation models on the organization's internal repositories — including proprietary frameworks, API definitions, test patterns, and code review comments. LoRA adapters make it efficient to create focused models: one adapter for a specific microservice architecture, another for the mobile codebase, a third for infrastructure-as-code templates. The resulting models suggest code that follows actual internal conventions, references real internal APIs, and respects the team's architectural decisions.

Deployment happens on-premise or within the organization's VPC using Ertas Cloud's private endpoint infrastructure. Models integrate with existing developer tools — VS Code, JetBrains IDEs, CI/CD pipelines — through standard API interfaces, providing code completion, review suggestions, and documentation generation without any source code leaving the network. Ertas Vault ensures that training data extracted from repositories is encrypted, access-controlled by team and project, and retained only as long as needed — giving security teams confidence that the AI pipeline meets the same controls as the codebase itself.

Key Features

Example Workflow

A fintech company with 200 engineers and a large Python/TypeScript monorepo wants to accelerate development while keeping its proprietary trading algorithms and risk models off third-party servers. The platform engineering team uses an internal script to extract 100,000 code-comment pairs, function signatures with docstrings, and pull request review comments from the monorepo, formatted as a JSONL dataset and uploaded to Ertas Vault. In Ertas Studio, the team selects a CodeLlama-13B base model from Hub and fine-tunes two LoRA adapters: one for Python backend code completion and another for TypeScript frontend patterns. Both adapters are deployed as private Cloud endpoints on the company's Kubernetes cluster, behind the corporate VPN. The endpoints integrate with VS Code through a custom extension that routes completion requests to the internal models. Within the first month, developers report that 40% of AI suggestions are accepted without modification — compared to 15% with the generic model — because the completions reference the correct internal packages, follow established error-handling patterns, and respect the team's TypeScript strict-mode conventions. All inference runs on company hardware, and the security team confirms via Vault's audit logs that no source code leaves the network perimeter.