Dzikri Syairozi

Lead Automation & AI Native Engineer

Horizon Analytics Labs — Remote · New York City, United States

AI Evaluation & RL Environments

• Lead benchmark and automation engineering at an NVIDIA Inception-backed lab that sells AI evaluations to frontier labs; primary author of the benchmark platform and of the audit tooling behind the company's public dataset Quality Index.
• Partnered with a Meta researcher to measure where the Muse Spark models break down on long, multi-step tasks, comparing them against Claude, Gemini, GPT, and GLM on one neutral harness with pass@k and agent-trajectory analysis.
• Designed 20+ long-horizon agentic benchmarks built to find where frontier models fail (radio-signal decoding, QUIC networking, LLVM backends, database recovery) in C, Go, Rust, Zig, and OCaml, each sealed in a sandbox and graded exactly, with no LLM judge.
• Built an automated tool that finds broken tasks in AI benchmark datasets — leaked answers, unsolvable tasks, and faulty graders — by pairing deterministic checks (file and leak-pattern scans, containerized do-nothing and reference-solution runs) with a non-deterministic classifier calibrated through RLCD (reinforcement learning for calibrated decisions), at under $0.001 of model spend per task.
• Audited datasets from 10+ publishers (OpenAI, Scale AI, Apple, Snowflake, Snorkel, Mercor): proved OpenAI's SWE-Lancer tasks cannot be passed even by their own reference solution, and found one image leaking the answers to all 726 tasks.
• Maintain and extend the company's enterprise AI floor-plan-to-3D platform for hotel and retail teams: Gemini-driven plan analysis, a WebGPU editor, Gaussian-splat and 360° tours, and a furniture and brand catalog in English and Chinese.

Founding Fullstack Engineer

Horizon Analytics Labs — Remote · New York City, United States

• Built the company's AI floor-plan-to-3D design platform from scratch: a vision-model pipeline turns floor plans into editable, photorealistic 3D rooms in a WebGPU/Three.js viewer, with GPU path-traced renders and Gaussian-splat room tours.
• Shipped a commercial AI content-generation platform — powered by a tool-calling, RAG, and dynamic-prompt agent system — for a Singaporean educational institution serving roughly 13,000 students.
• Deployed a microservices analytics platform for a chat application serving 500,000+ users on Kubernetes, and cut third-party production costs by about $2,400 a year by replacing Redis with a custom store.
• Designed a Domain-Driven Design ecosystem of 7+ microservices (REST + gRPC) with multi-tenant isolation and role-based access control.
• Shipped a Web3 trading platform and airdrop tasks automation in Next.js + TypeScript with RainbowKit/Wagmi multi-chain wallet support, on a PNPM-workspaces monorepo with a shared UI component library.
• Engineered "perpDexFarm," a market-neutral arbitrage bot that watches live price spreads across perpetual DEXs (Lighter, ApeX, Extended) and automatically opens hedged long/short positions with isolated-margin risk controls.

Software Engineer

OCT-PATH inc. — Remote · Tokyo, Japan

• Engineered an AI-powered enterprise document-management platform (Next.js, Supabase, AWS S3) with hierarchical file/folder organization, presigned uploads, and a configurable approval-routing engine featuring conditional multi-step routes, escalations, reminders, and full audit logging.
• Implemented a Retrieval-Augmented Generation pipeline for semantic document search plus an agentic AI assistant with OpenAI/Gemini tool-calling — file organization and submission/approval actions — over a streaming chat interface.
• Delivered admin and governance tooling (role-permission matrix, departments, positions, security-audit dashboards) plus analytics for KPIs, average lead time, and send-back rate, with English/Japanese i18n.

Software Engineer, AI Training Data

G2i Inc. — Remote · Florida, United States

• Evaluated and ranked AI-generated frontend code and UIs, enforcing rigorous standards for code quality, performance, accessibility (WCAG), and UI/UX fidelity to establish the ground-truth signal used to train next-generation AI coding models.
• Authored gold-standard training datasets in JavaScript, TypeScript, and React, and designed targeted prompts to refine LLM reasoning on user experience and interface generation.

Software Engineer

AlphaByte, Inc. — Remote · Tokyo, Japan

• Delivered AI-driven products for international clients (Japan, Ukraine, Singapore): an AI headshot/style image generator, a YouTube-to-podcast audio converter with summaries, and agentic features that auto-generate scripts, test scenarios, and equipment lists.
• Integrated machine-learning models into production and built the real-time data pipelines that serve inference for live product features.

Software Engineer

SPMB Jawa Timur — Surabaya, Indonesia

TypeScript, Next.js, React Hook Form, React Query, Cypress

• Maintained a high-traffic school-admissions platform serving 200,000+ students, with robust error handling for peak-load periods and Cypress end-to-end testing for stability during critical admission windows.
• Shipped student registration, proof generation, acceptance announcements, test-card-number generation, and geolocation-based proximity school assignment with cross-functional teams.

Software Engineer

Filinmovie — Surabaya, Indonesia

Next.js, Go, PostgreSQL, DRM, Video Transcoder, SSO

• Led a team building a Transactional Video-on-Demand (TVOD) platform, re-architecting it into micro-frontends across three sub-domains with DRM-protected streaming and video transcoding.

Institut Teknologi Sepuluh Nopember

Bachelor of Informatics Engineering — 3.71 GPA — Surabaya, East Java