RAG Specialist (Engineering Document Intelligence)

Job Specification RAG Specialist (Engineering Document Intelligence)

Department: MetisEI — Metis Knowledge

Reports to: Lead Developer, MetisEI

Location: Pune

Type: Full time, permanent

Purpose of the Role

MetisEI is CT Automotive's in-house AI-powered factory operating system and PLM platform. The Metis Knowledge stream ingests complex, highly structured automotive engineering and quality documents — PFMEA, DFMEA, Control Plans, Process Flows, KC lists, SIPs, SOPs, BOMs, ECRs and 8Ds — and serves them to AI agents and shop-floor systems where the retrieved content drives real production decisions and quality audits.

The RAG Specialist owns retrieval quality end to end. The defining challenge of this role: documents such as PFMEA and DFMEA are deeply structured — multi-level tables, row-and-column dependencies, severity/occurrence/detection linkages, cross-references between documents — and the data structure must be retained perfectly through ingestion, storage and retrieval. A retrieved PFMEA row that has lost its relationship to its failure mode, cause or control is not degraded information; it is wrong information, in a context where wrong information reaches a factory floor.

Key Responsibilities

• Design and own the ingestion pipeline for structured engineering documents: parsing, structure extraction and representation that preserves tables, hierarchies, cross-references and revision context with full fidelity.
• Own the retrieval architecture across the established hybrid design: whole-document context-window retrieval from Azure Blob for structured engineering documents, and embedded vector retrieval (Qdrant) for governance and knowledge content — including when each is used and how they combine.
• Guarantee structure-perfect retrieval: what is served to an agent or audit interface is verifiably faithful to the source document at the correct revision level, with no silent loss, flattening or reordering of structured content.
• Build retrieval evaluation as an engineering discipline: golden datasets from real CT documents, structural-fidelity checks, regression suites run on every pipeline change, and measurable retrieval quality metrics.
• Handle document diversity in the wild: inconsistent legacy formats, embedded images, merged cells, multi-language content (English/Spanish/Chinese/Turkish) — designing validation gates where automated extraction requires engineer sign-off.
• Support the cross-reference audit engine: retrieval that can answer linkage questions across documents (e.g. every high-severity PFMEA cause has a Control Plan line) reliably.
• Work with the lead developer and agent builders so retrieval contracts are explicit: defined input/output structures, null and missing-content behaviour, and revision-level guarantees.
• Continuously evaluate developments in RAG, structured retrieval, and long-context approaches, and bring justified improvements into production.

Essential Skills & Experience

• Proven production experience building RAG systems — not prototypes: ingestion, chunking/structuring strategy, embedding models, vector stores, hybrid search, re-ranking, evaluation.
• Demonstrable experience with structured and tabular document parsing where structure preservation mattered (engineering, legal, financial or regulatory documents), and able to explain how fidelity was verified.
• Strong Python; solid software engineering practice (testing, CI, versioned pipelines).
• Vector database experience (Qdrant preferred, or equivalent) and cloud object storage patterns (Azure preferred).
• Experience designing retrieval evaluation: building test sets, defining metrics, running regression.
• Comfort working with LLM-based systems and prompt/context engineering, including strict output contracts.

Desirable

• Familiarity with automotive quality documentation (FMEA per AIAG-VDA, Control Plans, PPAP, IATF 16949 context) — or demonstrable ability to learn a complex document domain quickly.
• Experience with document revision control / PLM concepts.
• Multi-language document processing experience.
• Exposure to safety- or compliance-critical information systems where retrieval errors have real consequences.

Personal Attributes

• Treats data fidelity as non-negotiable; instinctively distrusts unvalidated extraction.
• Pragmatic builder: ships working systems, then hardens them with tests.
• Clear communicator with non-specialists — engineers and quality teams must be able to trust and challenge the system.
• Comfortable in a fast-moving build environment where the specification evolves as the product proves out.