Skip to main content
Pakon Poomson
Public engineering evidence

Technical evidence across data and AI systems.

Six GitHub-backed case studies covering data platforms, RAG evaluation, ML/MLOps, multimodal extraction, Thai NLP governance, and public-data analytics.

AI Engineering

RAG systems, multimodal extraction, and Thai NLP governance workflows.

Demo availableRAG / AI Agent Systems

Customer Support RAG Triage Agent

Offline-capable RAG triage system for support tickets with grounded responses.

Role signal: Retrieval, ranking, guardrails, evaluation, and API contracts.

Customer support RAG triage dashboard and workflow.

What I built

  • System: A retrieval-grounded support triage workflow with observable steps, citations, evaluation, and fallback behavior.
  • Implementation: Typed workflow state keeps triage behavior inspectable.

What it proves

  • Skills: Retrieval, ranking, grounded generation, guardrails, offline evaluation, API contracts, and full-stack delivery.
  • Core stack: Python, FastAPI, RAG, LangGraph
RAGGroundingEvaluationGuardrails
DetailsGitHubDemo
Demo availableMultimodal AI Product

Receipt AI Expense Tracker

Privacy-aware Thai and English receipt extraction app with a local-first review workflow.

Role signal: Multimodal extraction, schema validation, provider routing, and IndexedDB.

Receipt expense tracker dashboard with totals, categories, and receipt history.

What I built

  • System: A multimodal receipt-to-expense workflow with schema validation, human review, local storage, and provider guardrails.
  • Implementation: Provider capability checks prevent images from reaching text-only routes.

What it proves

  • Skills: Multimodal extraction, Thai and English normalization, provider routing, local-first data design, and privacy-aware product engineering.
  • Core stack: Next.js, TypeScript, Zod, IndexedDB
Multimodal AIThai/EnglishLocal-firstValidation
DetailsGitHubDemo
Case studyThai NLP / ML Governance

Thai Review Sentiment Intelligence

Thai sentiment intelligence platform with governance, confidence routing, and a monitoring demo.

Role signal: Thai NLP, model governance, explainability metadata, and active-learning workflows.

What I built

  • System: A Thai sentiment workflow connecting inference, confidence-aware routing, human review, monitoring, and governance reports.
  • Implementation: Confidence routing sends uncertain predictions to human review.

What it proves

  • Skills: Thai NLP, ML governance, explainability metadata, monitoring, feedback queues, FastAPI, and React delivery.
  • Core stack: Python, scikit-learn, FastAPI, React
Thai NLPGovernanceMonitoringHuman review
DetailsGitHub

Data & ML

Local-first data platforms, reproducible forecasting, and public-data intelligence.

Case studyData Engineering / Analytics Platform

Urban Mobility Data Platform

Local-first data engineering and analytics platform for urban mobility datasets.

Role signal: Data pipelines, analytical modeling, API/dashboard delivery, and CI-safe reproducibility.

What I built

  • System: A local-first data platform connecting ingestion, validation, analytical modeling, APIs, and dashboard evidence.
  • Implementation: Deterministic sample pipeline keeps the reviewer path reproducible.

What it proves

  • Skills: Data pipelines, SQL modeling, reproducibility, API delivery, frontend integration, and CI-safe engineering.
  • Core stack: Python, DuckDB, dbt-style SQL, FastAPI
Data engineeringDuckDBAnalyticsLocal-first
DetailsGitHub
Case studyML/MLOps / Forecasting

Climate CO2 Forecasting ML

Local-first forecasting and MLOps demo with backtesting and interval monitoring.

Role signal: Time-series validation, model registry metadata, and experiment tracking.

CO2 forecasting dashboard with model comparison and atmospheric trend.

What I built

  • System: A forecasting workflow spanning validation, backtesting, interval monitoring, experiment evidence, serving, and visualization.
  • Implementation: Rolling-origin evaluation prevents future leakage.

What it proves

  • Skills: Time-series validation, model comparison, experiment tracking, model metadata, API delivery, and honest ML communication.
  • Core stack: Python, Forecasting, Backtesting, Intervals
ForecastingMLOpsBacktestingIntervals
DetailsGitHub
Demo availablePublic Data / Procurement Analytics

Thai Procurement Intelligence

Bilingual procurement intelligence platform using a governed 250-record official DGA/CGD snapshot.

Role signal: Public-data ingestion, provenance, bilingual evidence UI, and validation.

Thai procurement intelligence dashboard with bilingual public-data evidence.

What I built

  • System: A bilingual public-data analytics workflow spanning official-source ingestion, SHA-256 verification, provenance, validation, evidence views, quality reporting, and source-linked assistant retrieval.
  • Implementation: Production uses a governed 250-record official DGA/CGD snapshot with SHA-256 verification.

What it proves

  • Skills: Public-data ingestion, provenance design, bilingual UX, quality controls, security checks, and CI-safe analytics delivery.
  • Core stack: Python, FastAPI, Next.js, Data validation
Public dataProvenanceBilingualValidation
DetailsGitHubDemo
Case studyRetail Data Engineering / Quality Platform

RetailGuard Data Platform

Zero-cost local retail data platform with incremental Bronze extraction, protected Silver data, blocking quality checks, and DuckDB warehouse evidence.

Role signal: Incremental pipelines, PySpark, DuckDB warehousing, quality gates, and privacy-aware data engineering.

What I built

  • System: A local retail analytics platform spanning source seeding, incremental extraction, privacy-aware transformation, quality gates, warehouse loading, and evidence reporting.
  • Implementation: Default review path is fully local and requires no cloud account, billing account, free trial, or hosted service.

What it proves

  • Skills: Data engineering, PySpark transformations, warehouse modeling, privacy controls, quality gates, idempotency, Docker, and local-first reviewer workflows.
  • Core stack: Python, PySpark, DuckDB, PostgreSQL
Data engineeringPySparkDuckDBQuality gates
DetailsGitHub