Pilote Lite: an LLM assistant built into the product, shipped in three weeks

When every internal request needs a regulatory check, operations teams move at the pace of their legal experts. Nova Studio designed and built Pilote Lite, a business SaaS whose LLM assistant pre-analyses each case. The result: 70% less processing time, with no compromise on compliance.

In short

• The client: A large industrial group, legal and operations department
• The challenge: A bottleneck in the manual handling of complex requests
• The product: A business SaaS with a RAG-based LLM assistant, queried in plain language
• The delivery: The live product, its source code, documentation and two months of maintenance
• The impact: 70% less processing time, from 45 to under 15 minutes per case

---

1. The starting point: a scope narrowed to the real bottleneck

The operations department received a growing volume of internal requests requiring legal and technical expertise. Every request meant the same journey: find the reference documents in a vast library, interpret the applicable legal framework, then write a structured, compliant answer.

That manual process created delays of several days. Scoping ruled out replacing the expert from the start: the goal became giving them a copilot able to pre-analyse each case, leaving final validation to a human. That scoping decision is what made the project deliverable quickly, and acceptable to the legal teams.

2. Building: an architecture that makes invention impossible

On a regulatory subject, an AI that improvises is worse than useless. The whole architecture was designed to make invention impossible.

What the product is made of

• A RAG architecture. The model never answers from its general knowledge. It retrieves the applicable rule from the corpus provided by the company, then restates it.
• A vector database. The entire legal corpus and internal procedures are indexed so the model finds the right information in real time.
• An LLM through a secured API, for natural language understanding and synthesis.
• An interface built in Next.js, designed for field teams who are not lawyers.

In practice, Pilote Lite extracts the rule matching the situation described, applies it, and drafts an answer ready for review. The expert validates instead of searching.

3. At delivery: an autonomous, verified product

The SaaS went to production with its entire codebase, documentation covering the architecture, the technical decisions and how to take the project over, then two months of maintenance during real-world testing.

Measured results:

• Processing time: from 45 minutes to under 15 minutes per case, human review included.
• Reliability: by restricting the model to company documents only, the error rate was neutralised during testing.
• Adoption: teams singled out the clarity of the answers and direct access to legal information that had been siloed.

Do you have a SaaS project?

Pilote Lite shows that an AI assistant is only worth building into a product if it is constrained: trained on your documents, scoped to your business, and verifiable. That is what our OpenAI Select Partner certification guarantees.

We design and build your SaaS end to end, without you having to hire an engineering team. Let's talk about your project.