Proof over promises.
The challenge we were given, what shipped, and the measured results — not a description of how good it probably is.
Replacing three unsynchronized data stores with one engine
Challenge
A system needed fast exact lookups, priority-ranked retrieval, and ordered range queries over the same data, under a hard resource budget — without running three separate stores that drift out of sync.
Outcome
Solipher Index (built from this research) now covers all three access patterns from one engine with a fixed, predictable resource ceiling, validated on production-grade cloud hardware.
Results
- 1,407 of 1,414 real benchmark runs completed exactly as intended
- Throughput scaled from 54.8k to 1.83M ops/s across 1–64 concurrent threads
- Overtook common alternative approaches once real contention kicked in
Triaging 47,161 capsule-endoscopy frames under a fixed budget
Challenge
A wireless capsule endoscopy study produces 50,000–80,000 images per patient. The bottleneck wasn't classifier accuracy — it was retaining, ranking, and triaging that volume under real memory and storage limits with a defensible decision record.
Outcome
Solipher Triage (built from this research) now handles that triage layer, evaluated on the public Kvasir-Capsule dataset against six alternative approaches.
Results
- Scoring-model choice alone swung the flagged-for-review rate by more than 40 percentage points
- Full durability guarantees measured at 45.8–48.4% latency overhead, reduced substantially once batched
- A common alternative approach showed a 200–500× tail-latency spike invisible at the average