FrameSession Caching – Prior Art Disclosure and Reference Implementation

This repository documents and demonstrates FrameSession Caching, a method for persisting and reusing transformer inference state to enable efficient, deterministic semantic retrieval over large corpora.

It serves as both a prior art disclosure and a reference implementation, covering:

• Persistent serialization of key/value attention state and RNG metadata (".framesession" files)
• Hierarchical multi-pass query orchestration (HDM reasoning)
• Benchmark results across quantized transformer models
• Integration with a content management pipeline (ThoughtFrame.AI / eMediaDB)
• Experimental logs and corpus samples

---

🔍 Overview

FrameSession Caching enables transformer models to:

• Ingest documents once and reuse latent state across future queries
• Resume generation deterministically without re-prefill
• Perform hierarchical drill-down and coarse-to-fine relevance scoring
• Support low-latency semantic access to large, slow-changing corpora

Example domains:

• Legal documents, standards, textbooks
• DAM systems, enterprise knowledge bases
• Education and compliance workflows

---

📄 Included

• framesession-disclosure.md – Full technical disclosure (Markdown source)
• framesession-disclosure.pdf – Exported PDF version
• framesession-disclosure.sha256 – SHA-256 hash for verification
• src/ – Reference code modules (Java)
• logs/ – Semantic relevance test logs, model response timing
• corpus/physics_reference_chunk.pdf – Sample document used in evaluation

---

📜 License and Use

This work is released to the public domain as prior art. You are free to use, modify, and build on this work with attribution.

Please reference the original as:

Ian Miller, “FrameSession Caching for Efficient Semantic Retrieval with Quantized Transformers,” ThoughtFrame.AI, October 2025

No patents are being pursued. This is a contribution to the open AI community to prevent enclosure of foundational techniques in persistent transformer orchestration.

---

📫 Contact

Ian Miller, B.Eng (Computer), P.Eng, M.Sc (Technology Management)
Founder & Principal Architect
ThoughtFrame.AI – The Workflow Engine for Adaptive Intelligence
📧 [email protected]
🌐 https://thoughtframe.ai