Systems

Production infrastructure we built and operate. Every system includes real metrics and links to evidence.

ConstantSense Edge AI

Ten production models for phones, wearables, and microcontrollers. Heart screening, fall detection, industrial monitoring, document classification. Each benchmarked on approximately 100 real devices.

32 μs
Fastest inference (NPU)
1.01 μJ
Most efficient (LiteRT)
913
Real-device benchmarks
10
Production models
ONNX RuntimeCoreMLQualcomm AI HubINT8 Quantizationmodelsign

FinSense

Financial sentiment classification that runs on-device, plus a hosted API for teams. Scores headlines as bullish, bearish, or neutral in English, Spanish, Chinese, and Japanese — with an iOS app and a REST API for users.

88.2%
PhraseBank 50agree (held-out)
233 MB
INT8 ONNX (DeBERTa-v3)
42 μs
Inference (Snapdragon X2 Elite)
4
Languages
DeBERTa-v3Knowledge DistillationONNXREST APIiOS / CoreML

Constant Framework

504+ machine-checked mathematical theorems across three proof assistants. When software correctness matters — medical devices, financial logic, safety systems — we prove it.

504+
Verified theorems
3
Proof assistants
13/13
Radical function properties (Coq)
Lean 4Coq/MathCompIsabelle/HOL

Need something similar?

Every system here started with a conversation. Tell us yours.