Meet-Me: The Digital Twin

The Challenge

Professional portfolios are traditionally static, failing to capture the nuance of a Senior MLE’s decision-making process. I needed a way to provide technical recruiters and collaborators with an interactive, authoritative source of truth that could explain my architectural choices without manual intervention.

The Solution

I engineered Meet-Me, a localized RAG kernel that serves as my Digital Twin.

1. Contextual Brain (Vault Ingestion): The system indexes a curated Obsidian vault using Qdrant for vector storage. It leverages hierarchical tagging (Parent Categories + Retrieval Hooks) to ensure precise categorical alignment.
2. Inference Engine (Groq LPU): To achieve sub-second responsiveness, I utilized Groq’s Llama-3.3-70B models, optimized with prefix caching of static resume data.
3. Security & Rate Management: The backend is built on FastAPI, featuring a robust middleware stack including SlowAPI for rate limiting and custom API key validation to protect inference quotas.
4. Terminal Interface: The frontend is a custom Astro component designed with a terminal aesthetic, supporting Async Generators for real-time token streaming and an amber-themed high-contrast UI.

The Impact

Meet-Me successfully bridges the gap between high-level summaries and deep-dive technical specs. It maintains a zero-hallucination policy through strict cosine similarity thresholding and verified metadata flags, ensuring that every claim the Twin makes is grounded in my actual project documentation.