01
Memory
Memory is the missing layer.
RAG and agents that can’t remember context can’t ship to production. We open-sourced our take (`lakehouse-memory`) covering episodic, semantic, and working memory. We build the production extensions on top.
Context memory, built right on Databricks.
We implement memory-aware AI agents and RAG on your existing Databricks Lakehouse. Unity Catalog–native, Claude-optional.
The problem
Your RAG can’t remember the last turn.
Your agents lose state between runs.
Your vector DB is a parallel universe to Unity Catalog.
“To be effective in production, agents need state. They must remember what has already happened and resume work with full context intact.”
Databricks, A Hands-On Guide to Apps on Databricks (2026)
Three pillars
How we approach production AI on Databricks.
02
UC-native
Built where your data already lives.
Unity Catalog is the memory store. Governance, lineage, access control: you already have them. We use them, not a sidecar vector DB.
03
Claude-optional
Claude where it matters.
Bring Claude in for reasoning, planning, and evals. Keep Databricks-native models where they’re already strong. No vendor monoculture.
What an engagement looks like
Assess → Implement → Hand off.
01
Assess
2 weeks
We map your current Databricks workspace, identify the highest-leverage memory-shaped problem, and produce a written architecture and engagement plan.
02
Implement
6–12 weeks
We build it. Memory layer, agent runtime, RAG production pipelines, Claude integration, MLflow eval. Running in your workspace, owned by your team.
03
Hand off & advise
Ongoing, optional
We hand off documented, tested code and a trained team. Optional retainer for ongoing architecture review.
Credibility
Databricks Partner
Validated practice
Next Link Labs
Delivery partner
Book a 30-minute Databricks AI assessment.
We look at your workspace, what's stuck, and how memory-aware AI fits. No commitment.
Book with Travis →