AI Researcher & Software Engineer, currently open to remote opportunities.
- 🔬 I build and study ML systems end-to-end — from research (custom CUDA kernels, LLM architectures) to production-shaped full-stack apps.
- 📄 Portfolio, CV & writing: kronosdp.github.io
- 🤝 Open to collaborating on projects that put ML models in front of real users.
- NVIDIA — AI Research Intern (Jan-Jun 2025, Remote). Evaluated RWKV-v4/v7 against Transformers, LSTMs, and GRUs for formal regular-language expressivity; wrote a custom PyTorch CUDA C++ WKV operator for accelerated recurrent state transitions. → rwkv-research
- Meeting.ai — AI Engineer Intern (Jul-Dec 2025, Jakarta). Benchmarked ASR pipelines at 10+ simultaneous sessions to isolate latency bottlenecks; compressed a speaker verification model via knowledge distillation.
- Intergalactic Science Kingdom — AI & Backend Engineer (Jan-Jul 2026). Architected and deployed an LLM-backed product on Go/Django, integrating OpenRouter for multi-model routing, context management, and speech synthesis.
Full experience, education, and awards on my CV.
| Project | What it is |
|---|---|
| rwkv-research | From-scratch RWKV-7 implementation with a custom fused CUDA kernel, benchmarked against RNN/LSTM/GRU/Transformer on formal-language recognition tasks — grew out of my NVIDIA internship. |
| client-fit-vetting | Full-stack (FastAPI + React) MVP that standardizes recruiter-to-client candidate vetting with behavioral scoring and auto-generated coaching briefs. |
| Clickbait-Classifier-for-Gemastik-2023 | Clickbait detection for Indonesian headlines (IndoBERT vs. XGBoost/RF/Naive Bayes) — built for Gemastik, Indonesia's national student tech competition. |
| dicoding-permasalahan-insitusi-pendidikan | Student dropout prediction: EDA + a decision-tree classifier deployed behind a Streamlit scoring app. |
| bike-sharing-dashboard | Interactive Streamlit dashboard analyzing 2 years of bike-share rental data. |
