open to internship roles · malang, indonesia

Marvelino Husca

Software Developer, RAG Systems & QA-minded Builder

IT student who ships things and then tries to break them. I build AI-driven automation and retrieval systems, then spend as much time testing and hardening them as I did building them in the first place.

● Available for internship

Java · PHP (Laravel)

Python · RAG / LLM pipelines

Polinema — IT, 2nd semester

See Onboarding Buddy → Get in touch

01 — About

How I work

I'm an Information Technology student at Politeknik Negeri Malang (Polinema), with hands-on freelance and internship experience at Six Zenith Company, where I worked across Java, PHP (Laravel), and AI automation tooling.

My approach to building software leans heavily on testing discipline: isolating components, reproducing bugs before fixing them, and asking "why" before accepting a quick patch. That habit is what pulled me toward QA-adjacent work — I like systems that are provably correct, not just systems that happen to work today.

Outside of coursework, I'm building a portfolio of AI-automation projects, with Onboarding Buddy as the centerpiece — a retrieval-augmented chatbot I debugged, tuned, and rebuilt end to end.

02 — Skills

Toolbox

Languages

Java PHP Python JavaScript

Frameworks & Platforms

Laravel Langflow Vercel Serverless

AI / Data

RAG Pipelines AstraDB (Vector) Google Gemini Prompt Engineering

QA & Process

Test-case design Bug isolation Regression testing Agent eval metrics

03 — Featured Build

Onboarding Buddy

HR Onboarding RAG Chatbot

A retrieval-augmented chatbot that answers new-hire onboarding questions grounded in real company documents — built, broken, and rebuilt to be reliable.

● verified & live

Langflow AstraDB Google Gemini Vercel

buildReimplemented the Langflow pipeline as a Vercel serverless backend for a fully deployable live demo.

tuneTuned chunk size, overlap, and raised AstraDB search score threshold to 0.7 to cut noisy retrievals.

fixEnabled deduplication by file_path in the ingestion flow after catching duplicate chunks in retrieval.

guardAdded anti-fabrication instructions after catching the model hallucinating answers not in the source docs.

designBuilt a two-stage condensation architecture to give the chatbot reliable multi-turn memory.

testVerified against a four-category hero-query test set before shipping the live demo.

Try on → Source on GitHub

04 — Contact

Let's talk

Looking for QA-adjacent developer internships, and open to freelance work. Reach out — I usually reply within a day.

✉ E-mail GitHub ↗ LinkedIn ↗