Towards Persistent Case-Based Memory for Autonomous Data Science: A CBR-Augmented R&D-Agent with a Locally Deployable Small Language Model

2026-06-05 · Source: cs.SE updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Expert, extended

Summary

A novel CBR-augmented R&D-Agent integrates a persistent Case-Based Reasoning (CBR) layer into Microsoft's R&D-Agent framework, utilizing a custom Google AI Studio backend to deploy Gemma 4 31B Dense as its autonomous agent backbone. This system addresses the lack of persistent, cross-session memory in most top-performing agents and tests the viability of Small Language Models (SLMs) for local deployment. The CBR layer surgically overrides three R&D loop phases—hypothesis generation, code generation, and case retention—using structured case records with executable code snapshots, a five-gate quality filter, and a heuristic reuse-detection mechanism. Evaluated on two Kaggle competitions, Spaceship Titanic showed CBR achieving 0.8147 accuracy versus the baseline's 0.8098 with lower variance, while NOMAD saw the baseline achieve a larger SOTA gain. Heuristic reuse detection across 108 events revealed high semantic relevance (0.882 mean embedding similarity) and variable structural proximity (0.305 mean code-fingerprint similarity), suggesting conceptual guidance over direct code copying.

Key takeaway

For ML Engineers building autonomous data science agents in environments with limited GPU budgets or strict data privacy, you should consider adopting a CBR-augmented SLM architecture. This approach, exemplified by Gemma 4 31B Dense, provides persistent, quality-controlled memory, improving learning consistency and hypothesis quality. While it may trade some exploratory breadth for stability, it offers a transparent, locally deployable alternative to cloud-dependent frontier models, enhancing auditability and maintainability for your team.

Key insights

Persistent, quality-controlled case-based memory with a locally deployable SLM enhances autonomous data science agent performance and consistency.

Principles

• Structured case memory improves agent consistency.
• SLMs require engineering for structured output.
• Quality gates prevent knowledge base swamping.

Method

CBR integrates as a surgical subclass, overriding hypothesis, code generation, and retention phases. It uses prompt-based schema enforcement and JSON repair for SLM structured output, plus adaptive hang detection.

In practice

• Deploy Gemma 4 31B Dense for local agent backbones.
• Apply a five-gate filter for case base quality.
• Combine embedding and code-fingerprint similarity for reuse.

Topics

• Autonomous Agents
• Case-Based Reasoning
• Small Language Models
• Gemma 4 31B Dense
• Machine Learning Engineering
• Local Deployment
• Persistent Memory

Code references

• stofe94/cbr-rd-agent

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, AI Engineer

Related on AIssential

• Memp: Exploring Agent Procedural Memory — Procedural memory, dynamically built and updated, significantly enhances LLM agent performance and efficiency across diverse tasks.
• NEW GEMMA 4 beats GPT-5.4: The A4B Model — Gemma 4's 4B active MoE model excels in complex logical reasoning and self-correction, outperforming its 31B dense counterpart.
• Microsoft Foundry: Unlock Adaptive, Personalized Agents with User-Scoped Persistent Memory — User-scoped persistent memory enables adaptive enterprise AI agents while preserving data isolation and compliance.
• Comment remplacer ChatGPT par Claude (et pourquoi) — Claude's agentic AI model, with persistent memory and file interaction, offers a strategic advantage over traditional chatbots.
• SuperLocalMemory: Privacy-Preserving Multi-Agent Memory with Bayesian Trust Defense Against Memory Poisoning — Local-first AI memory with Bayesian trust and adaptive re-ranking defends against poisoning and personalizes retrieval without cloud or LLMs.
• Shanghai Jiao Tong University/DP Technology — Cognitive Accumulation and ML-Master 2.0 Architecture Analysis for Ultra-Long-Term Horizon Agent Science — Hierarchical cognitive caching enables ultra-long-horizon AI agents to maintain strategic consistency and efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.SE updates on arXiv.org.