DNA-inspired data-storage and AI-infrastructure research: a conversational data engine for AI models and databases, an embeddable encrypted key-value store for mobile and wearables, and a DNA-inspired compression engine for ML model weights. Three projects, one underlying idea — encode information the way biological systems do (four-symbol alphabets, content-addressed redundancy, density over raw speed) — applied to three different problems.
📖 Live site & documentation → · 🧬 DNA-inspired storage & quantum computing whitepaper →
A conversational data engine that stores AI models, images, documents, and ordinary database tables side by side — queried entirely in plain sentences, no SQL required. An LLM proposes an interpretation; a deterministic validator, never the LLM itself, decides what operation actually reaches storage. Tensor-level model access (retrieve one layer without touching the rest of the checkpoint), real tiered storage down to a working DNA-encoding archival tier (AES-256-GCM encrypted, then binary → A/C/G/T, with a real error-correction and corruption-simulation harness — see the whitepaper above), a bearer-token-gated REST API open for third-party NLU integration (RASA or custom).
cd mdc && pip install -r requirements.txt && mdc token issue my-integration && python -m mdc serveThe embeddable counterpart, for the places a server can't reach — a watch face, a phone app, a background service. Mobile/wearable-only: no CLI, no desktop app, no AI model storage (that's MDC's job) — just small generic objects. A 348 KB encrypted key-value store in Rust: XChaCha20-Poly1305 for every value and every key name, then DNA-encoded (the same A/C/G/T model as MDC's archive tier) before it touches disk, a 5-function C ABI, real cross-platform builds (macOS, Windows, Android — iOS/watchOS build from source, documented why; the macOS/Windows builds exist for desktop-companion-app embedding and a planned MDC Platform CLI bridge, not because mdc-lite is itself a desktop product).
MemCell — a DNA-inspired compression engine
A multi-stage lossless compression engine optimized for ML model storage, whose pipeline is explicitly modeled on DNA and cellular information storage — a pattern-genome codebook standing in for DNA's codon table, byte-plane transposition modeled on chromosome sorting, delta encoding modeled on neural adaptation. Self-contained C++17, zero external dependencies, reaching up to 88% reduction on repeated structures. See the whitepaper for how its biological framing maps to its actual pipeline stages.
Apache License 2.0 for MDC and mdc-lite (see each project's own LICENSE). See MEMCELL.md for MemCell's terms.