I'm an Applied AI Evaluation Specialist and Automation Architect with deep experience working on multimodal LLM behavior, safety, and workflow design. My background includes confidential work for major AI labs across audio-first models, general reasoning systems, and recommendation/shopping agents, where I specialized in stress-testing model behavior, identifying failure modes, designing complex prompts, and producing high-quality rubrics, annotations, and training signals.expertise spans multimodal evaluation (audio, text, vision/video entity tagging), persona and constraint prompt engineering, structured rubric design, and detailed reasoning analysis. I'm also comfortable working across automation platforms such as n8n, API integrations, and LLM-powered operational workflows, bringing a highly analytical, systems-oriented approach to problem solving.

Alec Wong

I'm an Applied AI Evaluation Specialist and Automation Architect with deep experience working on multimodal LLM behavior, safety, and workflow design. My background includes confidential work for major AI labs across audio-first models, general reasoning systems, and recommendation/shopping agents, where I specialized in stress-testing model behavior, identifying failure modes, designing complex prompts, and producing high-quality rubrics, annotations, and training signals.expertise spans multimodal evaluation (audio, text, vision/video entity tagging), persona and constraint prompt engineering, structured rubric design, and detailed reasoning analysis. I'm also comfortable working across automation platforms such as n8n, API integrations, and LLM-powered operational workflows, bringing a highly analytical, systems-oriented approach to problem solving.

Available to hire

I’m an Applied AI Evaluation Specialist and Automation Architect with deep experience working on multimodal LLM behavior, safety, and workflow design. My background includes confidential work for major AI labs across audio-first models, general reasoning systems, and recommendation/shopping agents, where I specialized in stress-testing model behavior, identifying failure modes, designing complex prompts, and producing high-quality rubrics, annotations, and training signals.expertise spans multimodal evaluation (audio, text, vision/video entity tagging), persona and constraint prompt engineering, structured rubric design, and detailed reasoning analysis.

I’m also comfortable working across automation platforms such as n8n, API integrations, and LLM-powered operational workflows, bringing a highly analytical, systems-oriented approach to problem solving.

See more

Language

Work Experience

AI Evaluation Specialist at Mercor Intelligence
August 1, 2025 - Present
Evaluate general-purpose multimodal LLM outputs (audio + text) across 18+ behavioral dimensions, including comprehension, prosody, persona, tone, reasoning, and instruction-following. Designed multi-turn adversarial prompts to stress-test model robustness in realistic conversations. Identified critical failures in reasoning, context retention, safety, and constraint adherence; documented causes and severities with detailed rationales. Implemented gold-standard rubrics (40–70+ criteria) for voice-first interactions and ensured alignment checks across domains.

Education

Coursework in Recreational Therapy at Florida International University
January 1, 2013 - December 31, 2016

Qualifications

Health & Life License (Licenses 2-15; incl. Annuities & Variable Contracts)
June 6, 2023 - December 1, 2026

Industry Experience

Software & Internet, Retail, Media & Entertainment, Education, Professional Services
    paper YouTube Water/Shark Shorts via n8n
    The YT Water/Shark Shorts workflow is a fully automated YouTube Shorts content generator that creates, renders, and publishes viral-style videos without any manual scripting, editing, or uploading. It runs entirely on schedule and transforms AI ideas into real YouTube content automatically. Here's how it works: The system triggers automatically on a schedule A Schedule Trigger kicks off the workflow at whatever frequency you choose (e.g., hourly, daily, weekly). No user interaction required. Fully autonomous. An AI agent generates a complete YouTube Shorts idea pack The Scripting AI Agent (powered by custom system instructions) generates: A YouTube Shorts title A fun, engaging description A detailed AI video-generation prompt All output is formatted in JSON and must follow strict storytelling rules for virality. The agent randomly rotates between three cinematic video types: Selfie Stick Ocean POV (Shark Encounters) Above-water, personal, dramatic. Giant Wave Catastrophe Massive wave approaching surfers or tourists. Aquarium Chaos Scene Orca viewing area with unexpected events. This ensures content variety and avoids repetition. An LLM validates the structure and fixes formatting A Structured Output Parser ensures the JSON is valid and production-safe before moving on. This prevents malformed prompts or broken uploads. A video-generation model creates the actual video Using the prompt, the workflow calls KIE AI’s Veo model, producing a vertical 9:16 Shorts-formatted video. The system handles: Rendering Authentication API calls Video retrieval Wait cycles while rendering completes All automatically. Once rendered, the system retrieves the final video The workflow polls the video record endpoint until the video file is ready. The video is uploaded directly to YouTube Through the Blotato API integration, the system: Uploads the AI-generated video Applies the generated title Applies the description Flags the video appropriately (not for kids) Posts it to the correct YouTube account This eliminates all manual uploading.
    paper InfiniteUGC Video Creator via n8n
    InfiniteUGC Video Creator is an end-to-end AI pipeline that turns a single Telegram message into a complete UGC-style video. It analyzes the reference image, generates accurate UGC prompts, creates multi-scene video scripts, renders each clip with AI, merges them, and sends the finished video back to the user fully automated and hands-free with zero manual effort. Here’s what it does: User sends an image + instructions through Telegram The workflow listens for a message via Telegram Bot and automatically extracts the attached photo and caption. This is the only input the user needs to provide. The system analyzes the image using GPT-4o The workflow uses a vision-language model to break down the image into structured details identifying brand elements, colors, characters, text, style, and more. This ensures the final video accurately represents the original image. An AI Agent generates UGC-style image prompts Based on the user’s instructions and the image analysis, an AI agent builds candid, authentic UGC photo prompts that match TikTok/Instagram creator aesthetics. This includes: natural lighting amateur iPhone-style framing authentic, casual settings accurate product rendering preserved text/logos The system generates a reference image using an external AI model A connected image-generation API (Fal AI) produces a high-quality image based on the UGC prompt. The workflow waits for the image to be ready and retrieves the file. The workflow automatically creates a multi-scene video script A second agent generates complete video prompts following strict UGC rules: realistic influencer-style recording casual dialogue about the product accurate depiction of the brand smooth transitions between scenes inferred video count based on requested duration Each scene is rendered individually using a video model (KIE AI Veo) The automation loops through each generated scene prompt and produces short 8-second clips that: feature the product match the requested style follow the AI-generated script use vertical or horizontal aspect ratios depending on user request All clips are merged into one complete UGC video The workflow automatically stitches the clips together, creating a smooth final asset ready for posting. The final AI-generated video is sent back to the user via Telegram No manual editing, prompting, or file handling required the user simply receives a polished UGC ad/video in a single message.