Computer science graduate from ITS. I build language-model systems, measure where they break, and explain the results to people who aren't engineers.
I research where language models fail, build the apps people actually use, and write so the rest of the team understands. Evidence before claims.
LLM pipelines, agentic verification over knowledge graphs, topic modeling, evaluation that reports its own limits.
TEEP research internship · CITI Lab, NTUST, Taipei
Three gates check every answer drawn from a Neo4j knowledge graph: is it faithful to its sources, are those sources trustworthy, and are they still current? If one fails, the agent retries.
0%
20-item ground-truth set, 3 runs
75%
median of 3 runs; low-trust category only 1/5 correct
57%
ablation; 100% with all gates removed
Undergraduate thesis · ITS
A hybrid pipeline: rule-based NER and temporal analysis build per-article constraints, three LLMs convert the article under those constraints, and rule-based validators plus entity injection repair what the models drop.
≤33% → 57–67%
480 articles per model; still below the 85% target
0.645–0.700
revised target ≥ 0.60 met by all three; the initial 0.70 target was reached only by GPT-4o-mini (0.700)
1,440
sampled from 26,526 valid articles
Text Mining final project · ITS
We scraped six national news portals, compared LDA, NMF and BERTopic, tuned BERTopic with Optuna, tested five embedding models, and shipped the result as a Streamlit dashboard for non-technical readers.
0.825
vs BERTopic 0.778 and LDA 0.616
3.5×
0.075 spread across embeddings vs +0.021 from Optuna
IT internship · UBS Gold, Surabaya
A cross-platform Flutter proof of concept that lets managers approve or reject internal requests (HR leave, finance verification, QMS) remotely, behind three authentication layers: registered device, app password and face recognition, with traffic tunnelled through a WireGuard VPN.
Community service project (KKN Abmas) · MCI Laboratory, ITS
A web-based financial management system for a kindergarten. We gathered requirements through surveys and interviews with teachers, treasurers and parents, then built five roles (administrator, secretary, treasurer, teacher, parent) covering enrolment, tuition, extracurricular payments, payroll and household expenses. Co-led.
Course project · ITS
A web app that recommends a study major: forward-chaining rules plus an XGBoost model behind a Flask REST API, with a Next.js front end.
PKM-K · national funding recipient
A multiplayer educational math game, funded through the national student entrepreneurship programme (PKM-K) and registered as copyright (HKI EC00202495300, Ministry of Law and Human Rights).
Course final project, written in journal-article format
The app translates American Sign Language (ASL) hand gestures into text in real-time by combining image processing and a YOLO-based object detection model.
Course final project, written in journal-article format (not published)
On an 83,561-row PaySim-style synthetic mobile-money dataset, compared class-weighted Logistic Regression and a Decision Tree, with SMOTE on the training set only.
0.769 / 0.862F1 / PR-AUC, Decision Tree
MCI Laboratory recruitment task
Two views on a sales dataset: monthly sales, which climb from roughly 550K in February to about 1.55M in December with a sharp dip in July, and order value by US state, led by California.
Machine learning course final project · written up on Medium
Labelled Play Store reviews with VADER, then trained Random Forest, SVM and Naive Bayes on those labels. VADER is an English lexicon and misreads many Indonesian reviews, so accuracy measures agreement with VADER, not true sentiment.
0.883Random Forest accuracy vs VADER labels
Course assignment · written up on Medium
Classification and clustering of how urban-development factors relate to a person's happiness.
0.70KNN accuracy (k = 3)
Deep Learning final project · ITS
Tested whether a small Transformer can forecast Microsoft's closing price, and whether an up/down framing works better. It could not: regression did worse than predicting the mean, and the classifier learned to always predict 'up'.
R² −2.83best regression run (S1)
0.556up/down accuracy
Center for IoT Innovation (CITI Lab), NTUST · Taipei, Taiwan
UBS Gold · Surabaya
Directorate of Business Cooperation and Management (DKPU), ITS · Surabaya
Information Intelligent Management Laboratory (MCI), ITS · Surabaya
Jawa Pos TV (Jawa Pos Update Surabaya) · Surabaya
Informatics Engineering Department, ITS · Surabaya
Most of what I know about getting things done with other people, I learned outside the classroom.
Computer-Informatics Engineering Student Association (HMTC) ITS
Helped coordinate a 120-member executive board through 20+ programmes in one term, and managed partnerships with companies, universities and government agencies. Previously General Secretary of Administration II (2024).
Schematics ITS (national IT event)
Led a multi-channel promotion strategy with 20+ media partners, tailoring the message for high-school students, university students and the public; the event reached 2,000+ participants.
Ikatan Mahasiswa Muhammadiyah (IMM) ITS
Prepared 15+ funding proposals under a tight budget and co-planned 10+ member-development and community-service programmes.
Forum Daerah Bengkulu ITS
Led 30 students from Bengkulu across ITS departments and ran routine programmes to build cooperation among members.
ITS
Treasurer, Keluarga Muslim Informatika (2024). Tim Kawal Gemastik ITS (2025), supporting ITS teams at the national ICT competition.
Master of Ceremony, IDSECCONF 2024 (Indonesia's annual IT security conference)
Workshop, keynote and panel sessions.
Speaker, Basic Media Schooling
Journalism training for students across departments.
Speaker on project and team management, Bluecamp, Informatics ITS
Technology
AI Fluency: Framework and Foundations
Anthropic Academy
Framework for working with generative AI effectively, ethically and safely.
Technology
Version Control System Training
Informatics Department, ITS
Git for collaborative code: branching, commits and versioning workflows.
Technology
Bootcamp and Seminar of Technology 2023
Informatics Department, ITS
IT career strategy and industry insights, with partners including Telkomsel and Dicoding.
Technology
Formulas and Functions in Microsoft Excel
amanexcel
Formulas vs functions, relative/absolute/mixed references, common errors.
Leadership
LKMM Tingkat Menengah XVII “Cakrawala”
ITS
Intermediate leadership and organisational management: integrity, growth mindset, planned change and collective leadership; completed with passing status.
Business
LKMW-TD (Student Entrepreneurship Training)
ITS Student Affairs with ITS Career
Entrepreneurial mindset, business planning and starting a venture.
Personal growth
ESQ Character Building
ESQ Training
Self-awareness and emotional and spiritual intelligence for academic and life challenges.
ITS News and Magazine (ITS Online), Public Communication Unit, ITS
Wrote and edited 150+ articles and magazine pieces, many of them explaining technical research to a general audience.
All articlesHave a role or a research idea? Let's talk.
Email is the fastest way to reach me.
Lathifah Sahda
Pick a subject — it opens in your mail app.