ARXIV:2604.21571 · LLM PERSONALIZATION · SUBMITTED 24 APR · 20:31 UTC · FRESHNESS STALE

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies

Chris Schneider · Philipp Schoenegger · Ben Bariach · arXiv

A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

Blocked on Code›Score4.0Evidence unverified

Opportunity summary

Pain A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

Evidence 0 refs | 3 sources | 50% coverage

Blocker Evidence unverified

Open Build Read PDF Signal Canvas Track

PROBLEM

A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning. This paper presents a three-layer architecture that decouples personal data from shared…

METHOD

Full abstract

Current model training approaches incorporate user information directly into shared weights, making individual data removal computationally infeasible without retraining. This paper presents a three-layer architecture that decouples personal data from shared weights by combining a static base model, composable domain-expert LoRA adapters that shape behavior without imparting user data, and per-user proxy artefacts whose deletion constitutes deterministic unlearning. Evaluation on Phi-3.5-mini and Llama-3.1-8B confirms per-user differentiation in which personal data influences outputs while remaining isolated, verified by a return to baseline after proxy removal (KL divergence of approximately 0.21 nats, 82-89% verification pass rate) and near-zero cross-user contamination. Because user-specific information never enters shared weights, the architecture mitigates model inversion, membership inference, and training-data extraction against shared model components by construction. The approach converts machine unlearning from an intractable weight-editing problem into a deterministic deletion operation that preserves personalization alongside privacy-enhancing guarantees and is compatible with differentially private stochastic gradient descent (DP-SGD) for privacy-preserving shared model improvement.

RESULT

ScienceToStartup currently rates this 4.0/10 on the public viability pass. The approach converts machine unlearning from an intractable weight-editing problem into a deterministic deletion operation that preserves personalization alongside privacy-enhancing guarantees and is compatible…

WHY NOW

LLM Personalization moved forward this cycle; last verified April 2026. Public score 4.0/10.

Continue into Read for claims, analysis, references, and neighboring papers.

Opportunity summary

Score4.0

PainA novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

Evidence0 refs | 3 sources | 50% coverage

Blockerno shell-level blocker reported

Analysis summary

A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

VerifiedSource: PDF linkedVerifiedPaperPack: citation fields availablePartialProof: unverified proof status

Competitive landscape

A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

Segment

LLM Personalization

Adoption evidence

No public code link in the paper record yet

Commercial read

4.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

{ "contract_version": "paper-r2", "paper_id": "a3e0e27c-e28d-46ec-816a-01e66179f8fc", "arxiv_id": "2604.21571", "canonical_route": "/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "active_tab": "synced from current hash by the drawer client", "selected_artifact": "separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "endpoints": { "paper_pack": "/api/v1/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p/paper-pack", "build_passport": "/api/v1/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p/build-passport", "mcp_resource": "sciencetostartup://surfaces/paper-workspace" } }

{ "surface": "paper", "mode": "paper", "query": "Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies", "normalized_query": "2604.21571", "route": "/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "paper_ref": "separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "topic_slug": null, "benchmark_ref": null, "dataset_ref": null }

{ "@context": "https://schema.org", "@graph": [ { "@type": "WebPage", "@id": "https://sciencetostartup.com/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p#webpage", "url": "https://sciencetostartup.com/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "name": "Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies", "description": "A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.", "isPartOf": { "@id": "https://sciencetostartup.com/#website" } }, { "@type": "ScholarlyArticle", "@id": "https://sciencetostartup.com/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p#scholarlyArticle", "headline": "Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies", "description": "A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.", "url": "https://sciencetostartup.com/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p", "sameAs": "https://arxiv.org/abs/2604.21571", "identifier": { "@type": "PropertyValue", "propertyID": "arXiv", "value": "2604.21571" }, "isAccessibleForFree": true, "isPartOf": { "@id": "https://sciencetostartup.com/#website" }, "datePublished": "2026-04-23T11:51:31.000Z", "author": [ { "@type": "Person", "name": "Chris Schneider" }, { "@type": "Person", "name": "Philipp Schoenegger" }, { "@type": "Person", "name": "Ben Bariach" } ], "additionalProperty": [ { "@type": "PropertyValue", "propertyID": "viabilityScore", "value": 4 }, { "@type": "PropertyValue", "propertyID": "researchDomain", "value": "LLM Personalization" } ] }, { "@type": "BreadcrumbList", "itemListElement": [ { "@type": "ListItem", "position": 1, "name": "Home", "item": "https://sciencetostartup.com" }, { "@type": "ListItem", "position": 2, "name": "LLM Personalization", "item": "https://sciencetostartup.com/topics" }, { "@type": "ListItem", "position": 3, "name": "Separable Expert Architecture: Toward Privacy-Preserving LLM", "item": "https://sciencetostartup.com/paper/separable-expert-architecture-toward-privacy-preserving-llm-personalization-via-composable-adapters-and-deletable-user-p" } ] } ] }

Competitive landscape

A novel LLM architecture decouples user data from shared weights using composable adapters and deletable proxies for privacy-preserving personalization and deterministic unlearning.

Segment

LLM Personalization

Adoption evidence

No public code link in the paper record yet

Commercial read

4.0/10 public viability

Direct

not classified

Adjacent

not classified

Substitute

not classified

Unknown

not classified

Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies

Separable Expert Architecture: Toward Privacy-Preserving LLM Personalization via Composable Adapters and Deletable User Proxies

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline

Claim map

Constellation map

Competitive landscape

Buzz

PDF

REFERENCES

Related Papers

Subscribe to the weekly brief

Build artifacts

Brief

Experiment plan

Validation checklist

Scientific founder

Translational engineer

Domain operator

GTM lead

Regulatory/clinical advisor

Timeline