Data Scientist · AI Researcher · Machine Learning · Retrieval Systems
Bryan Jericho.
>
Turning data into decisions. AI Researcher Intern in Taiwan. Researching federated vector search. Evaluating LLMs with LLM-as-a-judge. Building AI for stroke rehab.
I turn messy data into models that matter — from GoFood query recommendations at GoTo to federated vector search research in Taiwan.
Open to Data Scientist / ML Engineer rolesCurrently in Taiwan 🇹🇼 · Based in Indonesia 🇮🇩
0
prompts evaluated for the Sahabat AI Safety Benchmark
0
GoFood query recommendation effectiveness, up from 74%
0
university PDFs indexed for multimodal RAG
0
student developers in the community I led
01 — Experience
Where I've worked
AI research in Taiwan, data science at GoTo, and the developer community I led.
Oct 2026 – present · Chiayi, Taiwan · TEEPNow
AI Researcher Intern · National Chung Cheng University
Researching learning-based federated vector search: predicting per-client retrieval quotas and HNSW efSearch so a single round retrieves the global top-K across distributed clients, without moving raw data. Implemented FedAvg training and zero-shot onboarding for new clients in PyTorch and FAISS, with FedVS and FedOS as baselines.
Sept 2025 – Mar 2026 · Remote · Jakarta
Data Scientist Intern · GoTo Group
Lifted GoFood search query recommendation effectiveness from 74% to 92% by validating and labelling complex query data. Evaluated 300+ prompts for the Sahabat AI Safety Benchmark with an LLM-as-a-judge framework, and evaluated Dira, the customer-service chatbot, for safety, factual correctness and contextual relevance.
Aug 2025 – Aug 2026 · Makassar
Lead · Google Developer Group on Campus Unhas
Led a community of 900+ student developers and organised 30+ events, workshops and Study Jams, adding new learning tracks and industry partnerships.
Aug 2023 – 2027 (expected) · Makassar
Informatics Engineering · Hasanuddin University
GPA 3.83/4.00. Data scientist (evaluation) on the university's AI chatbot team: evaluated LLM responses for accuracy, relevance and safety, built Llama Guard-based safety checks, and ran data annotation and QA.
02 — Selected work
Things I've built
From pose tracking for stroke rehab to retrieval research. Open any card for the case study.
Featured · Undergraduate thesis
MoveAgain
A PWA for post-stroke rehabilitation: MediaPipe measures range of motion in real time, a Bayesian survival model predicts time to recovery, and an AI chatbot guides each exercise.
1,059,040 U.S. flights from Jan–Feb 2026: delays cascade through the day and are driven by carriers, not weather. Gradient Boosting flags delayed flights at ROC-AUC 0.65.
Post-stroke patients do most of their rehab at home, where nobody measures whether a movement is improving and therapists see progress only at the next visit.
02 · APPROACH
What I built
MediaPipe landmark detection turns a phone or laptop camera into a range-of-motion sensor for each guided exercise.
Functional PCA summarises each patient's range-of-motion curves; a Bayesian Weibull AFT survival model (brms) predicts time to recovery with credible intervals, served through an R Plumber API.
Evaluated with C-index, time-dependent Brier score, precision/recall and LOO posterior predictive checks.
03 · RESULT
What changed
Live at move-again.vercel.app. The model is trained and evaluated on simulated patient data; validating it on real clinical data is the next step.
30-second summary
Bryan Jericho — Data Scientist & AI Researcher
01Data science at GoTo. GoFood query recommendations from 74% to 92%; 300+ prompts evaluated for the Sahabat AI Safety Benchmark with LLM-as-a-judge.
02AI research in Taiwan. Learning-based federated vector search at National Chung Cheng University (TEEP), in PyTorch and FAISS.
03Builds. MoveAgain stroke-rehab AI, and a multimodal RAG over 214 university PDFs; 1st at DevForge Hackathon 2025.