dasch ai v1: Hybrid Token-Efficient Router

dasch ai v1 is an ultra-efficient, hybrid token-optimized routing agent engineered specifically to clear the hackathon accuracy gate while strictly minimizing API token usage. The system implements a rugged, production-ready three-tier architecture tailored for constrained computing environments. First, a zero-token deterministic keyword routing layer instantly categorizes incoming prompts into distinct capability domains. Second, the agent leverages an in-container, local 4-bit quantized Qwen-2.5-1.5B-Instruct engine running purely on CPU. This tier independently resolves standard text processing, sentiment analysis, and summarization tasks for absolute zero token cost, safely operating within the strict 4 GB RAM and 2 vCPU environment budget. Third, for highly complex reasoning, mathematical operations, or code generation tasks, dasch ai v1 securely routes requests to the lowest-cost option in the ALLOWED_MODELS tier via the Fireworks AI proxy. This fallback layer enforces hyper-strict parameter constraints, dropping the temperature to 0.0 for deterministic reliability and aggressively clamping max_tokens based on the category to eliminate conversational bloat and token bleed. This creates a ruthless balance between maximum task accuracy and minimum infrastructure cost.