The rapid proliferation of cloud-based large language models (LLMs) has introduced significant barriers including subscription costs, privacy concerns due to external API transmission, and operational dependencies on remote infrastructure. We present ECHOVIUM QPS V1.5 (Quantum Power Shell), a compact locally deployable LLM based on a 3B-scale open-weight instruction-tuned backbone, adapted for reasoning tasks using parameter-efficient fine-tuning and quantized deployment. QPS V1.5 achieves 64.2 ±1.3% accuracy on GPQA Diamond (PhD-level science reasoning) and 87.1 ±2.1% on LiveCodeBench code generation, demonstrating competitive performance relative to larger hosted models on selected reasoning benchmarks while maintaining strong efficiency among compact local models.Deployed via Ollama with 4-bit quantization, the model delivers up to 15 tokens/s while maintaining complete local data processing and zero marginal inference cost. Key contributions include a 7-stage QLoRA pipeline, reproducible benchmarking, backbone-agnostic framework, and production-ready deployment.