I build portfolio-grade, production-ready AI orchestration systems, deterministic RAG pipelines, and automated data architectures. My engineering philosophy centers on safety-first automation, multi-agent governance, and asynchronous performance—turning unpredictable LLM workflows into reliable enterprise solutions.
- Languages & Core: Python (Async), TypeScript, JavaScript, SQL, Google Apps Script
- AI & LLM Ops: Multi-Agent Orchestration, RAG Pipelines, Vector Search, Chain-of-Thought Auditing, LLM Evaluation, Hallucination Reduction
- Backend Frameworks: FastAPI, SQLAlchemy, Alembic, Pydantic, Uvicorn, httpx
- Frontend & UX: React, Vite, TailwindCSS, Shadcn/ui, Server-Sent Events (SSE)
- DevOps & Infrastructure: Docker, Docker Compose, GitHub Actions, Cloudflare Workers, Git
⚖️ TheArbiter — Multi-Agent Quality & Verification OS
A sophisticated multi-agent orchestration loop (Auditor, Architect, Tech/Logic Critics, Repair Loops, Final Verifier) designed to govern LLM outputs. Features multi-provider routing (OpenAI, Anthropic, Gemini, Groq, Ollama) optimized by performance/cost ratios, dynamic software architecture decomposition, and durable SQL persistence.
🔄 CVShift — Security-Focused Contextual RAG Platform
A full-stack SaaS platform utilizing ChromaDB vector search to ground LLM context and eliminate profile hallucinations during document tailoring. Built with FastAPI, React, TypeScript, and a deterministic ReportLab PDF generation engine. Implements Server-Sent Events (SSE) for streaming real-time agent execution progress.
⚓ InboxAnchor — Safety-First AI Inbox Triage Framework
An automated operations inbox triage system engineered around explicit safety verification categories and human-in-the-loop approval gates for destructive automation. Features modular AI agents, email provider abstractions (Gmail API/IMAP), alias routing via Cloudflare Workers, and full immutable audit logging.
- Enterprise Automation: Engineered custom SQL and Google Apps Script reporting architecture, reducing an operational 100-agent validation cycle from 40 minutes down to under 5 seconds (saving ~67 hours/month of manual enterprise overhead).
- Systems Thinking: Expert at translating loose operational policy frameworks into robust, deterministic programmatic guardrails.
📫 How to reach me: craciungabrielluca7@gmail.com | LinkedIn