Hi, I’m Ivan Golovach.

Built a receipt pipeline from OCR to human-verified item records, plus sealed household sync with authenticated bundles and deletion-safe replication.

Built one portable record model across iPhone and web, with uncertainty-aware analysis, clinician-readable exports, and separate StoreKit and Stripe/D1 entitlement paths.

Built policy-gated durable memory across text, voice, image, and web flows, with isolated Lambda canaries, runtime probes, and immutable rollback.

Maintain parsers for incompatible agent histories, normalized token and cache accounting, model-pricing fallbacks, and release reliability across terminal and web surfaces.

Maintain distributed model routing, staged split inference, llama.cpp integration, and failure recovery across changing peers and GPU runtimes.

Extracted sealed records, authenticated sync, deletion tombstones, and QR pairing into independently testable Swift packages while keeping private app data outside the public boundary.

Defined deterministic CBOR/COSE records with independent Rust and TypeScript implementations, then enforced byte-for-byte agreement through a strict conformance suite.

Built a Streamable HTTP MCP payment flow with pre-payment argument binding, idempotent retries, and signed receipts that prevent duplicate fulfillment.

Built deterministic simulations with explicit uncertainty and failure surfaces, plus checksum-guarded artifacts and an auditable evidence chain for reproducible results.

Built a versioned LLM research pipeline with cost gates, human approval, immutable run history, and strict separation between private evidence and public output.

Built a stateful mail execution engine with deterministic idempotency, layered send limits, human approval, reply and unsubscribe guards, and reconciliation for uncertain sends.