Data Analyst · Data Engineering · Data Governance
Full-time Data Analyst at an FMCG import company, where I own the data from source file to decision.
Background in geotechnical engineering; finishing a Master's in Applied Data Science with hands-on ML, NLP and LLM projects.
4 h → 35 min
ImpactFaster month-end reporting
Same checks for every source
ImpactConsistent numbers
Excel files → PostgreSQL
ImpactOne version of the truth
200k-row Excel → SQL
ImpactQuick, sure decisions
15 responsibilities in 4 areas: what I did and what changed. Open “Example” for a real case.
FMCG import company. I own the data end to end: extract it from source systems, transform it through Python data pipelines into PostgreSQL, and turn it into dashboards and analysis that answer business questions.
Designed the schema for sales, product, customer and stock data, then migrated it.
ImpactTrusted numbers for decisions: no sales tied to the wrong SKU, and new reports added without rebuilding.
One typed pipeline for every source, replacing per-file manual cleanup.
Impact~85% of processing time freed for analysis, with fewer errors reaching reports.
Python + Selenium scripts export portal data; Task Scheduler runs them on a schedule.
ImpactRecurring data and reports arrive on time without manual work.
Vendor PO PDFs → tables → Excel/CSV, on request.
ImpactLocked PDF records become analysable data, saving processing time per PO batch.
Desktop app that converts raw accounting data for the Express system, including Thai withholding-tax certificates (ภ.ง.ด. 3/3ก/53). Handles several source layouts.
ImpactSaves the finance team's time on every filing round: certificates and Revenue Department files come out correct and consistent in one run, so submitting on the RD website goes smoothly.
HTML dashboards (Python + PostgreSQL + Chart.js): brand, mall, salesperson, stock value, shipments.
ImpactSharper sales strategy, with self-service visibility for leadership anytime.
Sales by product, promotion, season, margin and mall.
ImpactDirectly informed pricing, product mix and new-import decisions.
Flags low stock for imports and items selling faster than their stock cover.
ImpactLower stockout risk and fewer lost sales.
Net sales reports, vendor forecasts and reconciled data on request.
ImpactSupports compliance and keeps vendor relationships on solid numbers.
Excel, Power BI and dashboards read PostgreSQL over ODBC; only the ETL can write.
ImpactEvery team works from the same fast data, and it stays intact.
Password levels on dashboards plus per-level database accounts.
ImpactSensitive figures never exposed to general staff or outsiders.
Checks that stop missing or unmapped keys reaching the database.
ImpactFewer downstream errors and less cleanup effort.
Fixed old records with wrong codes or names; scheduled PostgreSQL backups.
ImpactReliable history for trends and forecasting; business continuity protected.
Pre-test, tailored session, post-test for Excel, Power BI, dashboards and database access. First stop for data and IT help.
ImpactStaff work faster and more accurately; no escalation to other departments.
Shares PO ETA and stock status, places warehouse orders for sales, runs the staff sales channel.
ImpactField sales sell with current stock and ETA data; plus a small extra sales channel.
The yellow row is what I use every working day. The other tools come from my Master's and side projects.
Dashboards I run at work, rebuilt as demos, plus end-to-end work from my Master's and self-study. Every item links to its demo, repository or page.
The four documents behind my dashboards: data lineage from source file to dashboard, a data dictionary with owners and sensitivity classes, a role-by-dashboard access matrix, and 10 data-quality rules that block, warn or fix before load. Built on the same mock data as the demos.
Four of the dashboards I run at work, rebuilt as demos. Each one is a single HTML file generated by Python from PostgreSQL, with the data encrypted so it opens only with a password.
View full size
Ten years of sales in one page: YoY and MoM movement, best and weakest channel, customer and brand, full-year pace against target, and drill-downs by brand, channel, customer and category.
View full size
A briefing written for leadership: revenue against plan, gross margin against target, profit at risk, and the brands carrying the quarter. The page opens with a one-line summary of the quarter.
View full size
Built from the warehouse system export: product grade and damage share, lots expiring soon or already expired, and stock cover days per SKU, so the team knows what to sell or move first.
View full size
Puts a baht value on stock risk: total stock value, value at risk within 180 days, expired, slow movers, Grade B and damage, with value by brand and expiry ownership by company.
Logistic Regression, Decision Tree, KNN, SVM, Random Forest and AdaBoost compared end to end.
82–90% accuracyclassification_project ↗K-Means, Hierarchical and DBSCAN in Python and RapidMiner, validated with silhouette and adjusted Rand scores.
clustering_project ↗Regression modelling with a written R report.
regression_project ↗Book recommendations built with Spark MLlib, evaluated with precision@k.
recommendation1 ↗Hybrid recommender using all-MiniLM-L6-v2 embeddings and FAISS similarity search.
reccomd_project2 ↗PyTorch MLP predicting buy/hold/sell from MACD, RSI, Bollinger and ADX features.
85% test accuracyDL-4_SET50 ↗FAISS retrieval with a local Ollama model (gemma3:1b), Gradio UI and persistent context memory.
DL-3_RAG ↗Sentiment, NER, POS tagging and parsing with spaCy and TextBlob; FastAPI on AWS ECS Fargate with GitHub Actions CI/CD.
nlp1-spacy ↗TF-IDF features, classification and clustering in Python and RapidMiner.
98.35% accuracy (SVM)text_analytic_project ↗YOLOv8 + OpenCV on live feeds, fine-tuned for smoke detection.
Precision 0.986 · Recall 0.910 · mAP@0.5 0.965object_dectection_realtime_project ↗Tesseract OCR behind Flask, containerised with Docker on AWS EC2, deployed through GitHub Actions.
OCR1_project ↗CNN on MNIST (99% test accuracy), ResNet-50d, CLIP image search, GPT-2/BART generation and summarisation, OpenThaiGPT, Qwen chat. Branches dl1–dl10.
deep-_learning_project ↗Also: Gemini API chatbot with persistent history, and a LINE chatbot on the Claude API hosted free on Google Apps Script with token-budget alerts (private). Full index: github.com/JPP-J/Portfolio_Repository.
Database, ETL, dashboards and data governance for an FMCG importer. Details in Experience.
Led planning and supervision of soil and rock investigations for large-scale infrastructure projects (Department of Highways, MWA, NEDA) in Laos, Cambodia and Thailand. Coordinated multi-disciplinary teams and subcontractors, planned workloads, and presented results to clients. Managed project documentation and budgets, designed engineering structures, and worked with the cost-estimate team to cut construction costs by up to 25%.
Planned and coordinated field teams on road, building and river dam protection projects. Collected and analysed field data, standardised the reporting templates project managers used, and trained staff on procedures and new equipment.
Engineering taught me to distrust a number until I know where it came from. That habit is why my data work starts with validation and ends with access control, and why I'm heading toward data governance.
The fastest way to reach me is email: jidapa.pooljan@gmail.com
