Senior RAG & Knowledge Systems Engineer

The Enterprise; Hyderabad; On-site; INR 1,500,000 - 2,000,000; Full time

Job summary

The Enterprise in Hyderabad is seeking a Senior RAG & Knowledge Systems Engineer. In this hybrid full-time role, you'll design and build end-to-end retrieval systems and manage knowledge freshness to ensure data accuracy.

With 4-8 years in software or ML engineering and hands-on experience with vector databases, you will drive improvements in retrieval quality through evaluation and iteration. This role combines engineering and experimentation, perfect for those with strong Python skills and a passion for data systems.

Qualifications

• 4-8 years in software or ML engineering, 2+ years building production RAG systems.
• Deep hands-on experience with vector databases and embedding models.
• Experience tuning retrieval pipelines against real evaluation sets.

Responsibilities

• Design and ship end-to-end RAG pipelines for customer document types.
• Implement hybrid retrieval and manage knowledge freshness.
• Build evaluation harnesses to measure retrieval quality.

Skills

Python proficiency

Experience with vector databases

Knowledge graph construction

Familiarity with RAGAS

Tools

Pinecone

Weaviate

LangChain

Elasticsearch

Job description

Senior RAG & Knowledge Systems Engineer

Hyderabad, India / Hybrid Full-time mid level

Build retrieval systems that actually find the right thing, not just the closest vector, across messy, sprawling enterprise knowledge bases.

About the role

Most enterprise RAG failures aren't model failures, they're retrieval failures. Stale data, overlapping chunks, permission boundaries that get ignored, queries that retrieve plausible‑but‑wrong content. This role exists to close that gap. You'll design and own the knowledge layer for ArqAI's enterprise engagements: ingestion pipelines, chunking strategy, hybrid retrieval, reranking, evaluation harnesses, and freshness mechanisms that keep the system accurate as customer data changes. The work is equal parts engineering and experimentation; you'll measure what you build, iterate on real query sets, and explain why the retrieval broke and how to fix it. Strong opinions about embedding models, vector databases, and retrieval evaluation frameworks are a requirement, not a bonus.

What you'll do

• Design and ship end‑to‑end RAG pipelines: ingestion, chunking strategy, embedding selection, retrieval logic, reranking, and answer synthesis tuned to customer document types and query patterns.
• Implement hybrid retrieval (dense + sparse), metadata filtering, knowledge graph augmentation, and query decomposition where needed.
• Build evaluation harnesses to measure retrieval quality, recall, precision, faithfulness, and answer relevance; use them to drive iterative improvement, not just one‑off demos.
• Manage knowledge freshness: incremental ingestion pipelines, document versioning, and stale content detection so the system stays accurate as customer data changes.
• Integrate retrieval systems with customer identity and permissions layers so access‑controlled documents are only surfaced to authorised users.

What we're looking for

• 4–8 years in software or ML engineering, with 2+ years building production RAG systems or information retrieval applications.
• Deep hands‑on experience with at least one vector database (Pinecone, Weaviate, pgvector, Qdrant, or FAISS) and embedding models (OpenAI, Cohere, BGE, or similar).
• Strong Python proficiency. Fluent with LangChain, LlamaIndex, or Haystack, and clear about where each framework adds friction vs. value.
• Experience tuning retrieval pipelines against real evaluation sets, familiarity with RAGAS, TruLens, or equivalent evaluation frameworks.
• Comfortable working across document formats: PDFs, HTML, structured tables, transcripts, and semi‑structured enterprise data.

Nice To Have

• Experience with knowledge graph construction and graph‑augmented retrieval (Neo4j, Memgraph, or similar).
• Background in traditional information retrieval, search engineering, or Elasticsearch/OpenSearch at scale.
• Familiarity with ColBERT, SPLADE, or late‑interaction retrieval architectures.