Selective works, algorithmic research, and high-performance C infrastructure by Andy Curtis.
A selection of algorithms and systems I've developed over my career, largely in the public domain at this point.
- Quicksort Improvement (2019) — Designed an efficient modification to quicksort/introsort. By reusing the pivot sample to detect when input is already sorted, reverse-sorted, or all-equal, the algorithm finishes in O(n) with a single verification pass. Yields 5–30× speedups for sorted inputs with no regressions on adversarial cases.
- Ad Inventory Overlapping Set Problem (2011–2012) — Designed a real-time ad inventory model to handle millions of buys with overlapping targeting criteria using inverted indices and bitmap intersections. Enabled accurate forecasting and scalable allocation at web scale.
- Expected Frequency & Long Correlation (2009+) — An algorithm for search and recommendations leveraging entire user histories. By adjusting local co-occurrence with an expected vs. actual frequency correction, it systematically removes global popularity bias.
- Click-Based Search & Recommendations (2003–2005) — A framework leveraging full user sessions (queries + clicks) to learn correlations, improving ranking, personalization, and localization.
- Efficient Near-Shingling (2001) — An approach to near-duplicate detection that approximates shingling accuracy while reducing storage and compute costs. Documents are reduced to title + first + 10 longest sentence hashes, indexed in dual forward/inverted indices.
- EzResult Search Engine (1998–1999) — A distributed search engine written entirely in C/C++/assembly, independently utilizing inverted indices, tries, and cosine similarity. Supported instant index updates and was acquired in 1999.
A modular, multi-tier dependency graph of C libraries built for out-of-core data processing, search, and system reliability. Designed strictly for performance, simplicity, and composability.
|
|
Tools designed to tame the complexity of multi-repo ecosystems through declarative configuration and GitOps automation.
- scaffold-repo — A declarative fleet manager and build orchestrator. Resolves dynamic dependency graphs, enforces OSS license compliance, and automates Git branching/releases across dozens of interconnected micro-repos via a single Python CLI.
- scaffold-templates — The centralized Template Registry powering
scaffold-repo. Defines language stacks (C/CMake, Python), organizational profiles, and dynamic Jinja2 file routing to prevent vendor lock-in. Clone this to define your own fleet!