akiva ai

Open-source multi-agent crypto trading system with institutional-grade risk controls, multi-exchange support, and autonomous strategy execution.

Enterprise LLM proxy with unified multi-provider API, cost attribution, rate limiting, caching, and analytics -- based on LiteLLM.

Distributed LLM inference mesh for heterogeneous clusters -- pipeline-parallel sharding with SGLang, MLX-LM, and P2P networking. Fork of Parallax.

ML provenance and SBOM generator -- deterministic manifests with integrity verification and optional cryptographic signing for datasets, configs, code, and model weights.

Deterministic RAG evaluation toolkit -- retrieval metrics (recall, precision, MRR), corpus overlap detection, and CI regression gating without model calls.

Multimodal dataset QA -- directory scanning, file hashing, and exact duplicate detection for text, image, and audio datasets.

Data contract management and drift detection for ML/LLM pipelines -- automatic schema inference, validation, and statistical profiling with CI/CD integration.

Red-team and compliance test harness for LLM outputs -- detects PII leakage, secret exposure, and policy violations with CI-friendly reporting.

AI/ML evaluation harness -- versioned golden test suites, deterministic scoring, regression detection, and CI/CD-integrated reporting.

LLM inference cost and latency optimizer -- log analysis, SLO-based model recommendations, and routing policy simulation.