Loading content...
The conversational assistant and live Execution Inspector are currently in active development. In accordance with this repository's strict engineering discipline (AGENTS.md), the interface will be activated once the underlying ML router, Neon PostgreSQL persistence, and Qdrant RAG pipeline are fully built and verified.
When deployed, the assistant will answer technical questions regarding Mahad's experience, case studies, and engineering decisions using strictly cited sources. Every request will expose real-time routing confidence, latency budgets, and retrieval scores via the Execution Inspector.
Target latency < 5ms for intent classification, route prediction, and Roman Urdu detection without LLM API overhead.
Qdrant derived vector index paired with Neon PostgreSQL canonical chunks for 100% rebuildable, grounded retrieval.
Typed state graph with conditional retry boundaries and Server-Sent Events token streaming.
Circuit breakers, quota monitoring, and graceful fallback across permanent free-tier cloud resources.
Next.js 15, React 19, strict TypeScript, WCAG 2.1 AA accessibility, and automated CI tests.
Headless CMS schemas for projects, case studies, articles, and citations.
Relational source-of-truth, derived 384-d vector embeddings, and in-process ONNX router.
Typed state machine, Server-Sent Events live streaming, and Execution Inspector telemetry.