Alvin Hans

Selected Work

Selected work across search, analytics, and decision support.

Projects built on real operational data, with measurable quality and practical use in mind.

Search / Segmentation / Analytics / Decision Support

Selected WorkGitHub

Selected Work

Search & Retrieval

Hybrid Search Recommendation API

Built a hybrid retrieval API that combines lexical and semantic search into one usable product surface.

Proof: NDCG@10 0.798
->

BM25 + IndoBERT + FAISS gave a better balance of intent, relevance, and latency than a single search method.

Python /FAISS /PyTorch
Inspect Architecture
Customer Analytics

Customer Segmentation Analytics

Turned retail transaction data into segmentation and planning signals stakeholders could act on.

Proof: 8M+ tx / 2M+ customers
->

RFM features and clustering were the right balance of interpretability and scale for this use case.

Python /Scikit-Learn /SQL
See Results
NLP & Experimentation

Fintech Review ABSA

Turns Indonesian fintech app reviews into structured aspect-level signals for risk, trust, and service.

Proof: LoRA + baseline workflow
->

Three-aspect ABSA framing was more useful than a single polarity label because one review can surface multiple business issues at once.

Python /Transformers /Streamlit
View Evidence

Capability Snapshot

Retrieval Systems

Search Quality

Built a hybrid retrieval API that treats relevance, latency, and usability as one system problem.

BM25 + IndoBERT + FAISS, NDCG@10 0.798
Customer Analytics

Behavior atScale

Turned large-scale behavior data into segmentation and planning signals.

8M+ transactions, 2M+ customer profiles, 70K+ SKUs
BI & Reporting

Operational Visibility

Built reporting flows that made KPI monitoring clearer, faster, and more usable across departments.

Power BI across 4 departments, 100K+ records

NLP & Experimentation

Aspect-Level Signals

Built an end-to-end workflow for Indonesian fintech reviews that predicts risk, trust, and service instead of one coarse sentiment label.

NLP & Experimentation
Risk / Trust / Service, Baseline + PEFT, Streamlit

Experience

Data Science Intern

PT. Matahari Putra Prima Tbk | Sep 2025 - Feb 2026

Key Impact
  • ->Built customer segmentation on 8M+ transactions and 2M+ customers
  • ->Developed hybrid search system with NDCG@10 of 0.798
  • ->Extended retail intelligence workflows to 70K+ SKUs for demand-facing analysis
What I Did
  • Feature engineering, clustering, and retail analytics workflows
  • FastAPI-based search service with BM25, IndoBERT, and FAISS
  • Executive analytics and demand segmentation support

Data Visualization Intern

APP Group | Jan 2025 - Jul 2025

Key Impact
  • ->Built Power BI reporting suite across 4 departments
  • ->Queried and integrated 100K+ operational records
  • ->Reduced reporting friction through cleaner KPI delivery for internal stakeholders
What I Did
  • SQL-based reporting pipelines for operational dashboards
  • Cross-functional KPI reporting and dashboard logic alignment
  • Technical support for internal reporting systems

Certifications

IBM certification logo

IBM Data Engineering Professional Certificate

IBM In progress, expected Jun 2026

PASAS Institute certification logo

Certified International Specialist Data Modelling (CISDM)

PASAS Institute Issued Mar 2026

DQLab certification logo

DQLab Statistics using R for Data Science

DQLab Issued Feb 2023

About

I build production-ready ML and analytics systems for search, customer intelligence, and operational decision-making.

I care about measurable performance, clear architecture, and outputs that are easy to explain to both technical and non-technical teams.

The work in this portfolio is selected for scale, rigor, and business relevance rather than novelty alone.

Core Stack
Python / SQL / PyTorch / scikit-learn / FAISS / FastAPI / Power BI
Best Fit
Applied ML / Data Science / Analytics Engineering

What I Focus On

+
Retrieval systems: Balancing exact-match intent, semantic relevance, and production latency in one API surface.
+
Behavior analytics at scale: Turning 8M+ transactions into segmentation and planning signals stakeholders can use.
+
Operational decision support: Connecting analytical pipelines to reporting surfaces that teams can actually trust.
+
NLP methodology: Using normalization, baselines, and evaluation discipline instead of one-off model demos.
Final Notes

A focused portfolio ofapplied ML and data systems work.

Resume, project context, and contact details are below.

LinkedInEmail