Skip to content
View praysimanjuntak's full-sized avatar

Block or report praysimanjuntak

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
praysimanjuntak/README.md

An engineering notebook becomes a deployed museum robot, representing Pray Apostel's work in voice agents, edge ML, reasoning systems, and computer vision

portfolio · hugging face · linkedin · email

I build AI systems that have to work outside a notebook—real-time voice agents on robots, edge inference, reasoning models, and small products people can actually use.

A few things I've built:

  • 🧠 Qwen3.5-4B GRPO — post-training with verifiable rewards that lifted GSM8K pass@1 from 85.4% to 92.4%
  • 🔎 Deepfake Detection — an explainable image-forensics toolkit combining a learned classifier with ELA, FFT, noise, and EXIF signals
  • 📦 RAGBox — a local-first RAG command-line tool, packaged for pip install ragbox
  • 📉 Transformer Crypto — eleven experiments asking whether transformers can predict markets; the honest answer was more interesting than the optimistic one
  • 🧰 ToolsGratis — 15+ Indonesian-first utilities that process files entirely in the browser

Previously, I shipped a real-time voice agent with ~20 tools across 10+ tour-guide robots, and brought speech and vision models onto edge hardware. These days I'm exploring RL post-training, auditable multimodal systems, and software that earns a permanent browser tab.

Python · PyTorch · TypeScript · FastAPI · React · Docker · Jetson

I write up the useful parts—including the experiments that fail—on Substack.

Popular repositories Loading

  1. yolov12face yolov12face Public

    Python 2

  2. ragbox ragbox Public

    Python 1

  3. qwen3.5-4b-grpo-gsm8k qwen3.5-4b-grpo-gsm8k Public

    GRPO reasoning fine-tune of Qwen3.5-4B-Base on GSM8K with verifiable rewards (LoRA, vLLM rollouts). +7.0 pp pass@1 over base.

    Python 1

  4. htmlbeginner htmlbeginner Public

    HTML

  5. pokemonapi pokemonapi Public

    HTML

  6. dino-pinch dino-pinch Public

    JavaScript