Published Paper Details:

The exponential surge in digital data across distributed systems has intensified the demand for compression mechanisms that are both lossless and computationally intelligent. Traditional algorithms such as Huffman coding, Lempel-Ziv variants, and arithmetic coding have served the field well for decades, yet exhibit fundamental limitations when confronted with contextually rich or heterogeneous data streams. This paper presents an intelligent file compression and decompression system that harnesses DeepSeek-R1:1.5B -- a lightweight yet powerful large language model -- deployed locally via Ollama to drive context-aware probability estimation and entropy encoding. The proposed system integrates LLM-generated probability distributions into an arithmetic coding backend, enabling significantly improved compression ratios for text-dominant file types compared to classical baselines. Experimental evaluation across diverse datasets demonstrates measurable gains in compression efficiency while maintaining perfect lossless reconstruction. The system addresses practical deployment concerns through local inference via Ollama, mitigating privacy and latency overheads associated with cloud-based LLMs. This work bridges the theoretical potential of neural language models with tangible, deployable compression infrastructure and charts a course for hybrid AI-classical compression frameworks.