Khanoumi
Multi-agent RL pricing, AI content production, Text-to-SQL BI, and cloud MLOps for a 24,000+ SKU catalog.
Large-scale beauty & health e-commerce
System overview
Activate the visual to pause or resume animation.
Continue to project detailsCatalog sales, inventory and demand signals feed a multi-agent reinforcement learning pricing engine, with prices served through gRPC across 24,000+ SKUs. Separately, product attributes feed RAG and LLM content production, which became 60% faster for 10,000+ listings. Natural-language questions become SQL for business KPIs through LangChain.
Context
Khanoumi required production AI systems that could operate across catalog pricing, content generation, analytics, and deployment operations without slowing business teams down.
Problem space
Manual pricing review, slow PDP content workflows, and ad-hoc SQL queues were limiting scale across a large product catalog and non-technical stakeholder base.
Solution architecture
Architected a multi-agent reinforcement learning pricing engine using historical sales, inventory signals, demand elasticity, and low-latency gRPC serving.
Built an AI content system combining RAG-based attribute extraction, LLM prompt engineering, and image-synthesis orchestration for product visuals.
Delivered a natural-language-to-SQL BI platform with LangChain and custom prompts for direct stakeholder access to revenue, margin, churn, and conversion KPIs.
Owned Airflow, MLflow, GitLab CI/CD, Docker, Kubernetes, Terraform, and AWS Lambda/S3 release workflows.
- SKUs
- 24k+
Catalog covered by the pricing architecture.
- Faster PDP content
- 60%
Reduced production time for 10,000+ product listings.
- Model releases
- Same-day
Reduced release cycles from multi-day deploys.
Technology stack
- Python
- Pandas
- PostgreSQL
- LangChain
- gRPC
- Airflow
- MLflow
- Kubernetes
- Terraform
- AWS