Acropolis: A Portable, AI-Driven Semantic Search System for HPC
Abstract
LLM agents are now a primary interface to data repositories, but they locate files by issuing speculative regex queries through glob, grep, and file reads—a guess-and-check loop expensive in tool calls and tokens that ignores the metadata embedded in scientific file formats. We present Acropolis, a semantic search system that augments a tiered filesystem with adaptive-depth content indexing and exposes it to agents through a typed natural-language interface. Acropolis makes four contributions: a transparent file-summary operator that runs inside the I/O ingest path and is designed to amortize its dominant LLM cost through content-addressable idempotency; a pluggable search-backend abstraction surfacing lexical, vector, and graph engines as a deployment-time choice, evaluated empirically; an MCP server that exposes the resulting search surface to any compatible agent; and a scientific-data evaluation that quantifies the resulting agent-side savings on a 70-query CMIP6 retrieval benchmark, where an agent given Acropolis alongside its shell tools matches the shell-tool baseline on retrieval accuracy while substantially reducing both tool/backend calls per query and LLM token consumption.
Finding
On a 70-query CMIP6 benchmark over 149 climate-model files, an agent given Acropolis alongside its shell tools matches the shell-tool baseline on retrieval accuracy (70/70 versus 69/70) while issuing one-third fewer tool/backend calls per query (1.00 vs 1.50) and consuming roughly 3.1x fewer LLM tokens (3,574 vs 11,219).
System
- 2024–present
- 2017–present
More papers on IOWarp
Browse all papers on IOWarp- 2026
Trigger, Render, Reason: Multimodal LLM Agents for Science-Aware Scientific Simulation Monitoring
Hua Xu, Jaime Cernuda, Ana Gainaru, Scott Klasky, Anthony Kougkas, Xian-He Sun. International Conference for High Performance Computing, Networking, Storage, and Analysis (SC 2026)
- 2026
GPUH5: A GPU-first I/O Library for Scientific Data
J. Fernandes, Luke Logan, Xian-He Sun, Anthony Kougkas. 11th International Parallel Data Systems Workshop (PDSW 2026), held in conjunction with SC26
- 2026
Trigger-Gated Multimodal LLM Agents for In-situ Scientific Simulation Monitoring
Hua Xu, Jaime Cernuda, Anthony Kougkas, Ana Gainaru, Scott Klasky, Xian-He Sun et al. 11th International Parallel Data Systems Workshop (PDSW 2026), held in conjunction with SC26
- 2026
Type-Safe, Dynamically Defined Metadata Support for Scientific Applications
J. Teves, C. Brinker, Anthony Kougkas, Ana Gainaru, Scott Klasky, Xian-He Sun et al. 11th International Parallel Data Systems Workshop (PDSW 2026), held in conjunction with SC26
Citation
@inproceedings{pawar2026acropolis,
author = {Pawar, R. and Logan, L. and Cernuda, J. and Sun, X.-H. and Kougkas, A.},
title = {{Acropolis: A Portable, AI-Driven Semantic Search System for HPC}},
booktitle = {22nd IEEE International Conference on eScience (eScience 2026)},
year = {2026},
month = sep,
doi = {10.1109/eScience70237.2026.00045},
pages = {319--327},
url = {https://www.escience-conference.org/2026/papers},
publisher = {IEEE},
} Related notes
- 9 min
One destination, N tiers
Every tier supplies the application at once, and the application sees one destination. The thesis from a November 2026 talk to my group at the Gnosis Research Center, the six lines of I/O work behind it since 2016, and my bet that they are the foundation agents need for context.
- 4 min
SC26 in Chicago
What the group brings to SC26 in Chicago, the pressure from GPUs and agent fleets that runs through all of it, and an invitation to Thursday's Birds of a Feather on HDF5 in the age of AI-driven science.
- 6 min
CLIO as the entry point
CLIO should become the single place a developer starts with IOWarp, built as an application SDK with deployment profiles rather than an umbrella package or a gateway service. This is the direction for 2027; the API and the scope are still open.