Turning a static guide library into an autonomous AI that navigates, explains, and operates enterprise software.
Lemon Learning helps enterprise teams adopt software faster with in-app guides embedded in Salesforce. But as AI reshaped user expectations, static guides weren't enough. They needed AI that could understand the software, answer questions about it, and even operate it on behalf of the user.
Rakam built the complete AI layer. from a research-grade knowledge graph to an autonomous operator agent. turning Lemon Learning's product into one of the most advanced AI-powered learning platforms on the market.
Read our published research paper on ArXiv« We value Rakam's wide business vision and their development speed. We generated new revenue from AI features only three months after our first discussion. »
Pierre Leroux
CEO, Lemon Learning
3 mo.
To first AI revenue
96%
Operator precision
Thousands
Knowledge graph nodes
2+ yrs
Ongoing partnership
Business
Business Impact
Within three months of starting the engagement, Lemon Learning launched AI-powered features that became a new recurring revenue line. The AI layer transformed their commercial positioning. from a guide tool to an intelligent platform.
Quarterly AI roadmap sessions ensure continuous expansion. The published research paper on ArXiv strengthened both Rakam's and Lemon's credibility with enterprise buyers who need proof of technical depth.
Product
What We Built
GRAG. Graph-Based Retrieval-Augmented Generation
An automated pipeline that scrapes enterprise web applications (Salesforce) and converts them into state-action knowledge graphs with thousands of nodes. This powers grounded, context-aware AI assistance. no hallucinations, only real software knowledge.
AI Operator. Autonomous Guide Creator
An AI agent that autonomously navigates Salesforce, understands the interface through vision, and creates complete step-by-step Lemon Learning guides via their API. 96% precision. Creates a full guide in 6-7 minutes.
Streaming Chatbot
Real-time conversational assistant with dynamic knowledge retrieval from the GRAG system. Users get instant, accurate answers about any software feature.
Copilot & MCP Platform (in development)
Next-generation AI orchestrator with reasoning, tool calling, and planning. Will provide natural language access over Lemon's full API surface: analytics, processes, guide creation, back-office management.
Technical
Technical Architecture
The core innovation is the Graph-RAG architecture. a world-first approach that converts web applications into knowledge graphs for LLM-grounded assistance. We published this research on ArXiv.
The knowledge graph is built on Neo4j with semantic search via embeddings and Cypher queries. Response time: 10-15 seconds for complex graph traversals.
# Stack
Neo4j (knowledge graph, thousands of nodes)
Playwright (web scraping + vision)
CoreJS (DOM targeting)
AWS EC2 (operator, 10 workers)
MLflow (observability)
MCP servers (API abstraction)
# Performance
Operator precision: 96%
Guide creation: 6-7 min
Graph query: 10-15 sec
Move from AI hype to the real thing
Leave us a message, our team will get back to you within 24 hours.