🧠 AI / ML Research & Infrastructure

Building the Future of
Autonomous AI Systems

Independent research lab focused on LLM optimization, autonomous agents, and high-performance inference infrastructure on AMD hardware.

Get in Touch Learn More
22+
Platform Accounts
200+
GPU Hours/Month
5+
Active Projects
42
Cloud Credits Applied

About Us

AI/ML research and development team building production-grade autonomous systems.

🤖

Autonomous AI Agents

Building multi-agent systems for task execution, automation, and multi-step reasoning with tool use capabilities.

⚡

LLM Optimization

Fine-tuning and optimizing open-source LLMs (MiMo, DeepSeek, Qwen) for low-latency, cost-efficient inference on AMD hardware.

⛓️

Web3 Automation

AI-powered blockchain operations, smart contract analysis, on-chain data processing, and multi-chain automation.

🔧

MLOps Infrastructure

Full training pipelines with vLLM, TRL, Axolotl, Unsloth — from data curation to model serving and monitoring.

Technical Stack

Production-grade tools and frameworks powering our research.

🧠 AI / ML Frameworks

PyTorch Transformers TRL Axolotl Unsloth vLLM llama.cpp Outlines DSPy lm-eval-harness

🏗️ Infrastructure

AMD ROCm MI300X Docker Kubernetes Systemd Cloudflare Workers Nginx

💻 Languages & Data

Python 3.12 JavaScript Node.js SQLite PostgreSQL Redis Git GitHub Actions

Cloud Credit Use Cases

How we plan to utilize developer cloud credits for maximum research impact.

🏋️

LLM Fine-tuning & Evaluation

  • Fine-tune open-source models on custom datasets
  • Run benchmarks (MMLU, GSM8K, HumanEval)
  • RLHF / DPO / GRPO training experiments
  • Model merging and ablation studies
🚀

GPU Inference Serving

  • High-throughput inference via vLLM
  • Multi-model parallel serving for agents
  • Latency-sensitive workloads (<1s)
  • Structured output (JSON, regex, Pydantic)
⛏️

Mining Algorithm Optimization

  • RandomX CPU optimization (AMD EPYC)
  • GPU mining benchmarking (MI300X, MI250X)
  • Algorithm performance profiling
  • Power/efficiency tuning
🤖

Autonomous Agent Infrastructure

  • 24/7 agent execution (reasoning + tool use)
  • Browser automation, API orchestration
  • Cron-based background tasks
  • Multi-agent coordination

Expected Resource Usage

Monthly compute requirements for our active projects.

ResourceMonthly EstimatePrimary Workload
GPU Hours (A100 / MI300X)200 — 500 hoursLLM fine-tuning, inference serving
CPU Hours (EPYC / Xeon)1,000 — 2,000 hoursRandomX mining, data processing
Storage100 — 500 GBModel weights, datasets, checkpoints
Network Transfer50 — 200 GBAPI traffic, model downloads

Project Roadmap

6-month plan for research milestones and open-source contributions.

Q3 2026
Indonesian LLM Fine-tuning
Fine-tune MiMo v2.5 Pro on custom Indonesian language dataset for improved local language understanding.
Q3 2026
Multi-Model Inference Cluster
Deploy 3-5 models in parallel serving for autonomous agent workloads with automatic load balancing.
Q4 2026
Open-Source Agent Framework
Release autonomous agent framework with AMD GPU optimization, tool use, and multi-step reasoning.
Q4 2026
MI300X Benchmark Publication
Publish comparative benchmark results: AMD MI300X vs NVIDIA H100 for LLM inference workloads.
📊

Research Impact Goals

3+
Open-Source Releases
2+
Benchmark Papers
10+
Models Fine-tuned
100%
AMD ROCm Coverage

Cloud Credit Programs

Current applications and status across major cloud providers.

ProviderStatusCreditsAccounts
AMD DevCloud ✓ Submitted Pending approval 42 accounts
DigitalOcean ○ Applying $200 (60 days) 1 account
Google Cloud ○ Planned $300 free tier —
AWS ○ Planned Free tier —
Oracle Cloud ✕ Blocked Always Free Prepaid card rejected
IBM Cloud ✓ Active $200 (30 days) 1 account

Let's Collaborate

Open to research partnerships, cloud credit programs, and open-source collaboration.

📧 Contact

🔗 Links

  • GitHub
  • Semarang, Jawa Tengah, Indonesia