Skip to content
clariant @ arch : ~/clariant

~/clariant on ⎇ main []whoami --verbose

// output

Clariant

Benedictus

Tan.

AI engineer building multi-agent systems, RAG pipelines, and the unglamorous infrastructure that gets local LLMs to do useful work.

shipping @ dewaweb · internship · 2026

execute ·↑↓ history ·⌃l clear ·help for commands

$ tail ~/.now

  • Building AI infrastructure at Dewaweb.
  • Testing Hermes & OpenClaw agent harnesses.
  • Benchmarking Qwen 3.6 & Gemma 4 locally.

$ cat ~/focus.md

Multi-agent systems · RAG that holds up on real corpora · local LLMs.

I learn by building. The interesting part of AI isn't the model in isolation — it's the system you compose around it.

$ cat /etc/stack | shuf
PyTorch TensorFlow RAG n8n Chatwoot LangGraph Ollama Qwen Gemma Gemini Claude Code Next.js Astro Python TypeScript Multi-agent Vector DBs Supabase Vercel PyTorch TensorFlow RAG n8n Chatwoot LangGraph Ollama Qwen Gemma Gemini Claude Code Next.js Astro Python TypeScript Multi-agent Vector DBs Supabase Vercel

~/clariant on ⎇ main []ls -la ~/work/ --filter=featured

Shipping at Dewaweb.

Live internship — building the AI surface of an Indonesian hosting company. RAG support, agentic-tool research, infrastructure for an AI conference.

01. 2026
shipping

AI Copilot for Support Operations

RAG-grounded reply-suggestion copilot for the human support team — agents get LLM-suggested answers drawn from years of resolved tickets, while the human stays in control of every send.

RAG n8n Gemini Python
cat case-study.md
02. 2026
shipping

ScaleCon Surabaya 2026 — Technical Infrastructure

Owned the backend and interactive deployments for Dewaweb's AI-focused tech conference: subdomain provisioning, registration workflows, an AI Twibbon generator, and marketing automation.

Next.js n8n Unipile Google Sheets API
cat case-study.md
cd ~/work

~/clariant on ⎇ main []ls ~/projects/ --selected

Selected projects.

Personal and academic work — applied ML, CNN research on real datasets, design hand-offs to shipped products.

Piano Memory Match main menu — title, tagline, and Play / Free Play / High Scores / Settings buttons on a curtain-dark background
01. May 2026
shipped

Piano Memory Match

A music memory game — match pairs of notes by sight or by ear, across ten instruments. Built in Svelte + Tone.js with a pure-FSM game engine and 56 unit tests.

Svelte TypeScript Vite Tone.js
cat case-study.md
VISPA sign language learning platform — landing screen design
02. Feb – Jun 2025
shipped

VISPA — AI Sign Language Learning Platform

Full-stack web app for learning sign language, with a real-time random-forest hand-gesture detector under the hood. I owned the UI/UX in Figma and walked the design through implementation.

Figma Next.js React Flask
cat case-study.md
Traffic sign recognition CNN research — example classification output
03. Mar – Sep 2025
shipped

Evaluating Deep vs. Lightweight CNNs for Traffic Sign Recognition

Co-authored research paper comparing ResNet-50 against MobileNetV3-Large for ADAS traffic-sign recognition on GTSRB. Accepted at ISCT. I owned the ResNet-50 side end-to-end.

Python TensorFlow ResNet-50 MobileNetV3
cat case-study.md
cd ~/projects

~/clariant on ⎇ main []cat about.md

A short version.

I'm a Computer Science student specializing in Intelligent Systems, currently interning at Dewaweb on the AI side — RAG-grounded support, agentic-tool research, conference infrastructure.

Outside the internship I spend my time on multi-agent orchestration and getting local LLMs to do useful work. I learn by building.