A domain-specific ReAct RAG agent that accelerates onboarding in a legacy CMMS environment, encoding internal cases, schemas, and API docs into a corpus queryable in natural language. I built a 23-case evaluation set that exposed a silent failure no test caught — tool-argument casing mismatched indexed metadata, leaving 72 curated documents unretrievable. Fixing it raised retrieval pass rate from 25% to 85%.
"}},{"node":{"frontmatter":{"title":"SpeakCanada — AI English Learning Web App","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/23cdf22152de0575200628a6b3b8526d/11cb0/demo.jpg","srcSet":"/static/23cdf22152de0575200628a6b3b8526d/09491/demo.jpg 175w,\n/static/23cdf22152de0575200628a6b3b8526d/98dd4/demo.jpg 350w,\n/static/23cdf22152de0575200628a6b3b8526d/11cb0/demo.jpg 700w,\n/static/23cdf22152de0575200628a6b3b8526d/2ada4/demo.jpg 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/23cdf22152de0575200628a6b3b8526d/6916e/demo.avif 175w,\n/static/23cdf22152de0575200628a6b3b8526d/bb2f9/demo.avif 350w,\n/static/23cdf22152de0575200628a6b3b8526d/2e6d1/demo.avif 700w,\n/static/23cdf22152de0575200628a6b3b8526d/ace85/demo.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/23cdf22152de0575200628a6b3b8526d/f6b1d/demo.webp 175w,\n/static/23cdf22152de0575200628a6b3b8526d/1cb72/demo.webp 350w,\n/static/23cdf22152de0575200628a6b3b8526d/15050/demo.webp 700w,\n/static/23cdf22152de0575200628a6b3b8526d/86152/demo.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":298}}},"tech":["Next.js","TypeScript","Supabase","GPT-4o","Azure Speech","Vercel"],"github":"https://github.com/yfkiwi/speak-canada","external":"https://speak-canada.vercel.app/"},"html":"an AI Supported English & Culture Learning Platform with video, vocab, and AI-powered chatbot which help international students in Canada to practice everyday English in context-rich Canadian scenarios and get instant feedbacks.
"}},{"node":{"frontmatter":{"title":"Community Grant Assistant","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/6c5f4126effe7efee04b6236462780f6/7b23a/demo.png","srcSet":"/static/6c5f4126effe7efee04b6236462780f6/b9bb2/demo.png 175w,\n/static/6c5f4126effe7efee04b6236462780f6/aa464/demo.png 350w,\n/static/6c5f4126effe7efee04b6236462780f6/7b23a/demo.png 700w,\n/static/6c5f4126effe7efee04b6236462780f6/854f5/demo.png 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/6c5f4126effe7efee04b6236462780f6/c3835/demo.avif 175w,\n/static/6c5f4126effe7efee04b6236462780f6/cdba5/demo.avif 350w,\n/static/6c5f4126effe7efee04b6236462780f6/4fdae/demo.avif 700w,\n/static/6c5f4126effe7efee04b6236462780f6/2644e/demo.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/6c5f4126effe7efee04b6236462780f6/71830/demo.webp 175w,\n/static/6c5f4126effe7efee04b6236462780f6/b0cbb/demo.webp 350w,\n/static/6c5f4126effe7efee04b6236462780f6/d2119/demo.webp 700w,\n/static/6c5f4126effe7efee04b6236462780f6/10957/demo.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":332}}},"tech":["Next.js","TypeScript","FastAPI","Python","FAISS","ElevenLabs","GPT-4o","Azure"],"github":"https://github.com/yfkiwi/voicefirst","external":"https://orange-mud-07ee9a90f.3.azurestaticapps.net/"},"html":"Best Future Cities award-winning AI copilot that compresses community grant writing from weeks to hours. Built Next.js/TypeScript client with reusable workflows spanning intake, research, drafting, and submission review. Engineered FastAPI backend with FAISS-powered RAG to map local documents and policy libraries to grant requirements. Delivered voice-first flows via ElevenLabs TTS/STT and GPT-4o personas, orchestrated through Azure Functions and CI/CD.
"}},{"node":{"frontmatter":{"title":"Snowy Weather Domain Adaptation for Monocular Depth Estimation","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/6025014bb50174d39f5430f2a5fc3512/51cd7/demo.jpg","srcSet":"/static/6025014bb50174d39f5430f2a5fc3512/3ea88/demo.jpg 175w,\n/static/6025014bb50174d39f5430f2a5fc3512/230e3/demo.jpg 350w,\n/static/6025014bb50174d39f5430f2a5fc3512/51cd7/demo.jpg 700w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/6025014bb50174d39f5430f2a5fc3512/19d98/demo.avif 175w,\n/static/6025014bb50174d39f5430f2a5fc3512/b52d4/demo.avif 350w,\n/static/6025014bb50174d39f5430f2a5fc3512/0ddc0/demo.avif 700w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/6025014bb50174d39f5430f2a5fc3512/eac4f/demo.webp 175w,\n/static/6025014bb50174d39f5430f2a5fc3512/3eafe/demo.webp 350w,\n/static/6025014bb50174d39f5430f2a5fc3512/f8597/demo.webp 700w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":351}}},"tech":["Python","PyTorch","Stable Diffusion","ControlNet","DepthAnything","Teacher-Student Model"],"github":"https://github.com/yfkiwi/snow_dataset_preparing","external":""},"html":"How do you help self-driving cars “see” in a snowstorm? I built a synthetic-snow pipeline with ControlNet + Stable Diffusion and used a teacher–student model to transfer depth knowledge from clear weather to snow, improving depth estimates in snowy scenes.
"}},{"node":{"frontmatter":{"title":"Tims Cup Detection","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/a3e1cd0ac465238b710aeaacf9656523/91ccc/demo.jpg","srcSet":"/static/a3e1cd0ac465238b710aeaacf9656523/b5fea/demo.jpg 175w,\n/static/a3e1cd0ac465238b710aeaacf9656523/ddb3d/demo.jpg 350w,\n/static/a3e1cd0ac465238b710aeaacf9656523/91ccc/demo.jpg 700w,\n/static/a3e1cd0ac465238b710aeaacf9656523/c0336/demo.jpg 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/a3e1cd0ac465238b710aeaacf9656523/78482/demo.avif 175w,\n/static/a3e1cd0ac465238b710aeaacf9656523/e5e45/demo.avif 350w,\n/static/a3e1cd0ac465238b710aeaacf9656523/f4066/demo.avif 700w,\n/static/a3e1cd0ac465238b710aeaacf9656523/09d73/demo.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/a3e1cd0ac465238b710aeaacf9656523/98832/demo.webp 175w,\n/static/a3e1cd0ac465238b710aeaacf9656523/bccff/demo.webp 350w,\n/static/a3e1cd0ac465238b710aeaacf9656523/5d47d/demo.webp 700w,\n/static/a3e1cd0ac465238b710aeaacf9656523/36b56/demo.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":426}}},"tech":["YOLOv8","FastAPI","Docker","React","Vite","ShadCN UI","TailwindCSS","Roboflow","FiftyOne"],"github":"https://github.com/yfkiwi/timhortons-cup-detection","external":"https://huggingface.co/spaces/Eviekiwi/timhortons-cup-yolo-detection"},"html":"Is that cup a real Tims? This YOLOv8 model says yes or no by recognizing the classic red-and-white design and filtering out other coffee cups. Try it and see if your photo passes the Canadian test.
"}},{"node":{"frontmatter":{"title":"BloomWatch","cover":{"childImageSharp":{"gatsbyImageData":{"layout":"constrained","placeholder":{"fallback":"data:image/png;base64,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"},"images":{"fallback":{"src":"/static/56420907341d6b1f14d71cc9680bc947/d884a/demo.png","srcSet":"/static/56420907341d6b1f14d71cc9680bc947/3a042/demo.png 175w,\n/static/56420907341d6b1f14d71cc9680bc947/90d1f/demo.png 350w,\n/static/56420907341d6b1f14d71cc9680bc947/d884a/demo.png 700w,\n/static/56420907341d6b1f14d71cc9680bc947/346a3/demo.png 1400w","sizes":"(min-width: 700px) 700px, 100vw"},"sources":[{"srcSet":"/static/56420907341d6b1f14d71cc9680bc947/ab4e6/demo.avif 175w,\n/static/56420907341d6b1f14d71cc9680bc947/024b9/demo.avif 350w,\n/static/56420907341d6b1f14d71cc9680bc947/4713d/demo.avif 700w,\n/static/56420907341d6b1f14d71cc9680bc947/06a22/demo.avif 1400w","type":"image/avif","sizes":"(min-width: 700px) 700px, 100vw"},{"srcSet":"/static/56420907341d6b1f14d71cc9680bc947/32753/demo.webp 175w,\n/static/56420907341d6b1f14d71cc9680bc947/ab163/demo.webp 350w,\n/static/56420907341d6b1f14d71cc9680bc947/8c302/demo.webp 700w,\n/static/56420907341d6b1f14d71cc9680bc947/bc096/demo.webp 1400w","type":"image/webp","sizes":"(min-width: 700px) 700px, 100vw"}]},"width":700,"height":343}}},"tech":["React","JavaScript","Google Earth Engine","NASA MODIS","ERA5","Mapbox GL","Python"],"github":"https://github.com/yfkiwi/bloomwatch","external":"https://bloomwatch-ten.vercel.app/"},"html":"NASA Space Apps 2024 Local Winner & Global Nominee
\nEver heard about a legendary superbloom... three days after it peaked? BloomWatch predicts these rare events using 20 years of NASA data, giving you the exact 10 day window to witness millions of wildflowers carpeting California's deserts
"}}]}}}