You will get LLM Response Evaluation for Indonesian Language and Culture
Project details
I evaluate LLM responses for accuracy, tone, and cultural fit in Indonesian, drawing on real professional experience auditing AI outputs for factual and cultural alignment. As a native speaker, I catch nuances a non-native evaluator or automated system would miss. I bring the same rigor to this work as I do to my technical ML projects: consistent scoring criteria and honest, careful evaluation rather than rushing through responses.
What's included
| Service Tiers |
Starter
$50
|
Standard
$120
|
Advanced
$220
|
|---|---|---|---|
| Delivery Time | 3 days | 5 days | 10 days |
Number of Revisions | 0 | 0 | 0 |
Frequently asked questions
About Wincent
AI/ML Engineer | End-to-End Machine Learning Systems
Medan, Indonesia - 10:56 pm local time
I build machine learning systems from the ground up, starting with messy, real-world data and ending with something that actually works in production. That means cleaning and preprocessing data, engineering the right features, training models in Python and TensorFlow, and tracking every experiment properly so you know exactly what worked, what didn't, and why. I don't just hand off a model file and disappear, I document the process, explain the tradeoffs, and make sure you understand what you're actually getting.
What sets me apart from most ML freelancers is my background. Before moving into AI engineering, I spent 2 years doing design work for 10+ brands. That means I can also make your results understandable and presentable to the people on your team who aren't going to read a confusion matrix for fun, whether that's a founder, a stakeholder, or an investor.
📊 What I've built:
- Achieved 88% test accuracy (89% training) predicting lung cancer risk from 5,000 records using KNN (k=20), published in JUMINTAL (DOI: 10.55123/jumintal.v4i1.5460) and integrated into the LungHealth mobile app
- Automated end-to-end retraining and deployment of a clinical Random Forest model (303 records, 14 features) via GitHub Actions, tracking 9 hyperparameter configurations in MLflow and containerizing via Docker
- Developed a 6-class image classifier on 14,000+ images using EfficientNetB0, achieving 94.96% training (93.96% test) accuracy, compiled into SavedModel, TF-Lite, and TFJS formats
Implemented ML pipelines and web applications using Python, TensorFlow, scikit-learn, MLflow, Docker, FastAPI, React, and TypeScript
⚙️ How I work:
Get into the data first. Before writing a line of model code, I figure out what's actually predictable from what you've got, and I'll tell you honestly if something isn't feasible
Build it properly. Preprocessing, feature engineering, and training, done in Python with scikit-learn or TensorFlow, no shortcuts
Track everything. Every experiment logged in MLflow, every deployment containerized in Docker, so nothing is a black box you have to trust blindly
✅ Why work with me:
I won't oversell what a model can do. If your data can't support what you're hoping for, I'll tell you early, not after you've paid for something that quietly underdelivers. I'd rather set realistic expectations upfront than have you find out the hard way.
Not sure if your data can support what you need? That's exactly the kind of question I like to dig into!
Steps for completing your project
After purchasing the project, send requirements so Wincent can start the project.
Delivery time starts when Wincent receives requirements from you.
Wincent works on your project following the steps below.
Revisions may occur after the delivery date.
Review prompts and responses
I'll go through each prompt-response pair to understand context before scoring.
Evaluate against criteria
Score each response for factual accuracy, tone, and cultural fit (and safety, for Standard/Advanced tiers).