FRAGATA: Semantic Retrieval of HPC Support Tickets via Hybrid RAG over 20 Years of Request Tracker History

2026-04-16 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Cloud Computing & IT Infrastructure · Depth: Advanced, long

Summary

Fragata is a semantic ticket search system developed and deployed at the Galician Supercomputing Center (CESGA) to enhance knowledge reuse from over twenty years of Request Tracker (RT) history. The system addresses RT 4.4.1's limitations, such as poor indexing, case-sensitivity, and lack of semantic understanding, by combining dense retrieval using 384-dimensional embeddings from the paraphrase-multilingual-MiniLM-L12-v2 model with classical BM25 lexical retrieval. It also incorporates query-aware reranking via the mmarco-mMiniLMv2-L12-H384-v1 cross-encoder and supports incremental updates without service interruption. Fragata's architecture offloads expensive ingestion stages to the FinisTerrae III supercomputer and integrates external documentation, providing robust search capabilities across Spanish, English, and Galician queries, including those with typos or morphological variants.

Key takeaway

For AI Scientists developing knowledge retrieval systems for technical support, Fragata demonstrates a robust approach to overcoming limitations of legacy ticketing systems. You should consider a hybrid RAG architecture that integrates dense and lexical retrieval, query-aware reranking, and incremental ingestion with hot-swap capabilities. This strategy allows for efficient semantic search over large, heterogeneous, and multilingual historical data, significantly improving knowledge reuse and reducing resolution times.

Key insights

Hybrid RAG systems can effectively transform decades of unstructured support ticket data into a semantically searchable knowledge base.

Principles

• Combine dense and lexical retrieval for robust search.
• Use reranking to refine initial search results.
• Ensure continuous availability with hot-swap mechanisms.

Method

Fragata uses SQL extraction, normalization, and chunking of RT history, then applies hybrid retrieval (FAISS for dense, BM25 for lexical) with Weighted Reciprocal Rank Fusion (WRRF), followed by cross-encoder reranking and domain-specific score adjustments.

In practice

• Implement query variants for multilingual and typo-tolerant search.
• Offload compute-intensive indexing to HPC resources.
• Employ transactional watermarks for reliable incremental updates.

Topics

• Retrieval-Augmented Generation
• HPC Support Tickets
• Semantic Search
• Hybrid Information Retrieval
• Request Tracker

Code references

• facebookresearch/faiss

Best for: AI Scientist, MLOps Engineer, AI Engineer, Research Scientist

Related on AIssential

• FRAGATA: Semantic Retrieval of HPC Support Tickets via Hybrid RAG over 20 Years of Request Tracker History — Fragata enhances HPC support ticket retrieval using hybrid RAG over two decades of historical data.
• The retrieval rebuild: Why hybrid retrieval intent tripled as enterprise RAG programs hit the scale wall — Enterprise RAG is shifting from simple vector search to complex hybrid retrieval for production agentic AI.
• RARE: Redundancy-Aware Retrieval Evaluation Framework for High-Similarity Corpora — Existing RAG benchmarks misrepresent real-world performance due to high document redundancy and similarity in enterprise corpora.
• Your RAG System Isn’t Retrieving. It’s Guessing. — Current Retrieval Augmented Generation (RAG) systems largely rely on vector databases and the assumption that semantic similarity equates to relevance, a principle powering billions in infrastructure.
• Rethinking Enterprise Search: How Cortex Search Turns Data into Business Impact — Hybrid retrieval combines keyword and semantic search to enhance enterprise data accessibility and AI application grounding.
• Benchmarking RAG Architectures Locally on a Real Financial PDF — Local RAG on complex financial PDFs requires specialized extraction and architectural choices to achieve meaningful accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.