Freelance Agent & RAG Pipeline Engineer

About The Role: We are building an enterprise-grade agentic RAG pipeline that processes and retrieves knowledge from large, multilingual document corpora — spanning Arabic and English — across multiple business domains. The system must meet strict cross-domain isolation and safety requirements. As the pipeline engineer, you will own the full stack: from raw document ingestion and OCR through embedding, vector storage, and retrieval, all the way up to the agent loop that orchestrates tool calls, manages confidence, and triggers escalation when needed. You will decide how documents are chunked, how the embedding model is enforced consistently, and how the Qdrant collections are structured per team. You will also shape the prompting strategy that governs multilingual behavior and enterprise tone, and put in place the guardrails that ensure responses stay within domain boundaries. This is a hands-on, end-to-end role for an engineer who is comfortable operating at every layer of an AI retrieval system and takes full ownership of quality — from retrieval precision to hallucination prevention. Key responsibilities:  Implement ingestion pipeline: OCR (PaddleOCR), text cleanup, chunking strategy, metadata tagging.  Implement embedding pipeline and enforce embedding model consistency  Implement Qdrant schema design, per-team collections, snapshot export/import, and update strategy (delta vs full replace).  Improve retrieval quality (k selection, reranking if needed, context window management). • Implement agent loop (e.g., LangGraph) and tool-calling patterns (RAG, OCR, escalation, summarization). • Implement escalation triggers: o Model self-assessment (confidence) o Rule-based classifier for “must escalate” queries • Ensure responses respect guardrails and do not leak cross-domain information. • Tune prompts for multilingual behavior (Arabic/English) and enterprise tone. Required Skills & Experience: • Practical RAG engineering, text processing, vector DB operations (Qdrant). • Performance optimization on CPU-constrained devices. • Strong testing mindset for retrieval quality and isolation. • Strong LLM prompting, tool-use design, agent frameworks (LangGraph/LangChainequivalent). • Experience with evaluation methods (golden sets, regression tests for hallucinations/leakage). • Understanding of safety/guardrails in enterprise contexts. Nice to Have (Optional): • Familiarity with PaddleOCR or similar open-source OCR tooling. • Experience deploying AI workloads on edge or CPU-only servers. • Prior work with multilingual NLP (Arabic/English tokenization, normalization). • Knowledge of Qdrant internals: payload filtering, HNSW tuning, snapshots. • Exposure to LangGraph or comparable stateful agent orchestration frameworks. • Contribution to or maintenance of evaluation harnesses / golden-set test suites.