Training + evaluation
Dataset admission, secret rejection, distributed parallelism, checkpoint identity, held-out evaluation and independent promotion gates.
I build the full path from model-facing infrastructure to a reliable user-visible result: training and evaluation systems, retrieval and agents, authenticated APIs/SDKs, MCP and WebMCP-oriented browser-tool boundaries, streaming React/Next.js interfaces, Node/Python services, PostgreSQL, Rust runtime components, Solidity contracts, CI/release systems, and an experimental programming language/compiler/runtime.
Public evidence is neutralized and sanitized: employers can inspect engineering depth without private product names, credentials, endpoints or business identifiers.
Dataset admission, secret rejection, distributed parallelism, checkpoint identity, held-out evaluation and independent promotion gates.
Content addressing, safe corpus paths, model routing, tools, durable jobs, evidence receipts, replay and recovery.
React/Next.js workspaces, SSE state, responsive drawers/sheets, keyboard/focus behavior, mobile states and accessibility.
Authenticated contracts, deterministic discovery, operation identity, bounded paging, idempotency and receipt-based outcomes.
Node/Python services, PostgreSQL repositories, journals, transactional outboxes, workers, safe retries and reconciliation.
Lexing, parsing, resolution, typing/effects, HIR/MIR verification, lowering, bytecode/VM and Rust bootstrap runtime work.
Solidity, wallet handoff, nonces/replay protection, signature checks, exact settlement, finality and fail-closed behavior.
Docker, CI/CD, artifact identity, observability, release gates, security boundaries, rollback and runtime verification.
These demos run server-side on the deployed edge worker. They are bounded and deterministic rather than pretending an external model call happened.
Token-scored retrieval across this portfolio's public capability corpus.
Ready.
Typed planning stages based on the task's intelligence, protocol, UI and durability needs.
Ready.
Lexer demo: identifiers, integers, strings, operators and punctuation with positions.
Ready.
Verifies that the public backend is executing and reports its bounded API surface.
Ready.
The live source routes below are sanitized public samples reconstructed from implementation patterns I have actually built.
admit -> run -> checkpoint -> held-out eval -> independent review -> promotion candidate
Secret rejection, distributed execution configuration and evidence-vs-acceptance separation.
View sourceauthenticate(actor) tools/list -> catalogHash tools/call(operationId) -> receipt
Canonical discovery, stable mutation identity, paging and fail-closed operation receipts.
View sourcemeta -> delta* -> done | error request identity bound terminal state authoritative
Explicit stream identity and terminal-state correctness.
View sourceoperationId -> request hash -> receipt same transaction -> durable outbox unknown -> reconcile before retry
Idempotent mutation identity and durable asynchronous work.
View sourcelex -> parse -> resolve -> typecheck -> HIR/MIR -> verify -> lower -> bytecode -> VM
Language front end, typed IR verification and bounded execution contracts.
View sourceintent + payer -> replay guard allowlisted asset -> transfer received == requested
Non-reentrancy, replay protection and exact settlement.
View sourceAdditional sanitized examples cover Rust safety boundaries, browser-tool integration, SDK clients and accessible interface mechanics.
validate path + bytes scan bounded input return findings
Fail-closed validation for credential/key/executable risk.
View sourcesession + scope operation identity context sanitization
Browser-tool architecture with scoped capability checks and secret-field removal.
View sourcehealth() search(query) ask(question)
Small SDK with JSON contracts and bounded error handling.
View sourceopen -> focus Tab trapped close -> restore
Escape handling, focus trapping, ARIA labeling, scroll locking and previous-focus restoration.
View sourceStrongest day-to-day footprint: TypeScript/JavaScript and Python; systems and trust-critical work in Rust, Solidity and SQL/PostgreSQL.
AI Engineer - Applied AI Engineer - AI Platform / Developer Experience Engineer - Full-Stack AI Engineer - Agent Systems Engineer - Full-Stack Software Engineer - Developer Tooling Engineer - Smart Contract Engineer.
Hi — ask about Jemuel’s AI/ML work, API/SDK/MCP systems, frontend, backend, compiler/runtime engineering, Solidity, production hardening, roles or contact details.