Systems ObservatoryProject evidence passportView ML Infrastructure track

C++17 + PythonMLI

annlite

HNSW with SIMD kernels, Python bindings, and FAISS comparison.

51 tests; at 98.5% recall it performs 26x fewer distance computations than exact scan.

Inspect source
annlite project overview
Verified result1.83xFAISS QPS at 0.999 recall

Evidence chain

The claim, with its attack surface exposed.

Recruiters can scan the result. Engineers can inspect how it was produced and where it stops being valid.

  1. Problem

    Approximate retrieval claims are meaningless without the full recall-versus-latency curve.

  2. Mechanism

    Layered HNSW graph, diversity-aware neighbor selection, SIMD distance kernels, and GIL-free search.

  3. Attack

    Exact ground truth, recall-matched FAISS comparison, and thread-count invariance.

  4. Boundary

    FAISS remains 10-20% faster in the low-recall regime; the crossover appears only at high recall.

Reproduce it

One command to the test boundary.

The repository contains the implementation, tests, benchmark harness, and documented limitations behind this page.

git clone https://github.com/asp53826/annlite && cd annlite && pip install -r requirements-dev.txt && make test
Verified withexact recallFAISS baselineconcurrency invariants