Archipelia
Retail | ERP
A complete turnkey AI engagement. built, delivered, and handed over with full code passation.
Archipelia needed an AI layer for their retail ERP. a chatbot for user support, intelligent search across their documentation, and data obfuscation to protect sensitive customer information. They also needed the system to be fully transferable to their internal team.
Rakam built the complete system using the Rakam Systems framework, with spaCy NER for data obfuscation, pgvector for semantic search, and multi-model support. The engagement ended with a full code passation. infrastructure, documentation, and deployment pipelines included.
« Archipelia's AI system was built, delivered, and handed over. a complete turnkey engagement from code to infrastructure. »
Archipelia
Client partnership
Complete
Code handover
spaCy
NER obfuscation
4
Models supported
BI Agent
Potential identified
Business
Business Impact
Archipelia received a fully operational AI system with complete code ownership. The turnkey delivery model means their internal team can maintain, extend, and deploy the system independently. no vendor lock-in, no ongoing dependency.
The data obfuscation layer using spaCy NER ensures sensitive customer data is automatically masked before reaching the LLM, addressing data privacy concerns that are critical for retail ERP systems handling PII and financial information.
Product
What We Built
AI Chatbot & Search
Conversational AI assistant with semantic search across Archipelia's documentation and knowledge base. Users ask questions in natural language and get contextual, sourced answers about ERP features, workflows, and configurations.
Data Obfuscation (spaCy NER)
Automatic détection and masking of personally identifiable information using spaCy's Named Entity Recognition. Customer names, addresses, financial data, and other sensitive fields are obfuscated before any LLM processing.
Multi-Model Support
Architecture supporting 4 different LLM models, allowing Archipelia to switch between providers based on cost, performance, or availability. Future-proofs the system against model deprecation or pricing changes.
Technical
Technical Architecture
Built on the Rakam Systems framework, the system uses pgvector for semantic search and spaCy NER for real-time data obfuscation. Multi-model support across 4 LLM providers ensures flexibility and cost optimization.
The entire system was delivered via GitHub Container Registry with full CI/CD pipelines, infrastructure-as-code, and documentation. ready for Archipelia's team to operate independently on Kubernetes OVH.
# Stack
Rakam Systems (agent framework)
pgvector (semantic search)
spaCy NER (data obfuscation)
MLflow (observability)
GitHub Container Registry
Kubernetes OVH (orchestration)
# Delivery
4 LLM models supported
Complete code passation
CI/CD + IaC included
BI agent potential identified
Move from AI hype to the real thing
Leave us a message, our team will get back to you within 24 hours.