PROJECTS
Selected Projects
Personal Home Server
Designed and deployed a self-hosted infrastructure using a Synology NAS to run applications, manage storage, host websites, and provide secure remote access.
Website Analytics & Performance Analysis
Analyzed website data to understand visitor behavior, identify trends, examine document downloads and look for ways to improve website performance and bandwidth usage.
Receipt to Inventory & Analytics
Designing a self-hosted workflow that turns receipts into structured data for household inventory and spending analysis using local automation, AI extraction and PostgreSQL.
CASE STUDY
Personal Home Server
Why I Built It
I wanted to better understand how the services I use every day actually work behind the scenes. Building my own server gave me an opportunity to learn by setting up, configuring and troubleshooting the infrastructure myself.
Objective
I wanted to build my own home server to centralize my files, applications and services while developing my knowledge of networking, servers and self-hosting.
The Challenge
I wanted to access my services remotely while keeping my NAS and home network as secure as possible. I also wanted to run several applications without exposing every service directly to the Internet.
Architecture
The Solution
I configured a Synology NAS as the central server and used Docker to deploy and manage applications. I also configured reverse proxy services, Dynamic DNS, SSL certificates and Tailscale for remote access.
Technologies Used
The Result
I now have a centralized home server capable of running multiple services while giving me greater control over my data, applications and remote access.
What I Learned
This project gave me practical experience with Docker, networking, DNS, reverse proxies, SSL/TLS, remote access and server administration. It also improved my ability to troubleshoot technical problems and understand how different services work together.
Skills Demonstrated
Docker applications running on the Synology NAS.
Key Takeaways
- Built and managed a self-hosted environment from the ground up.
- Gained practical experience with Docker and server administration.
- Learned how DNS, HTTPS and reverse proxies work together.
- Developed troubleshooting skills by solving real configuration problems.
CASE STUDY
Website Analytics & Performance Analysis
Why I Did It
I wanted to analyze the data coming from the website to give the people responsible for it a better understanding of how it was being used. By looking at visitor trends, interactions, document downloads and bandwidth usage, I wanted to identify what was working, where there was room for improvement, and how the website could better respond to the needs of its visitors.
Objective
The goal was to use the available website data to better understand visitor behavior, identify important trends and provide useful information that could help improve the website. I also wanted to look at document downloads and bandwidth usage to understand how visitors were using the site's resources and find ways to reduce unnecessary bandwidth consumption.
Questions I Wanted to Answer
Visitor Behavior
- How has website traffic changed over time?
- Where are visitors coming from?
- How are visitors interacting with the website?
- Are there noticeable changes in visitor behavior over time?
- Which periods experience the highest activity?
Document Downloads
- Which types of documents are downloaded the most?
- Which documents appear to be the most requested by visitors?
- Do document downloads contribute significantly to bandwidth usage?
Bandwidth & Performance
- How much bandwidth is being used?
- What is contributing most to bandwidth consumption?
- Are there ways to reduce bandwidth usage without negatively affecting visitors?
- Could frequently downloaded or large documents be optimized?
- Which documents may be contributing the most to bandwidth consumption?
Data & Tools
The analysis was primarily carried out using Google Sheets to organize the website data, calculate trends, compare periods and create visualizations. Microsoft Excel was also used as part of the analysis.
Key Metrics
Website Traffic Over Time
Website traffic showed an overall upward trend during the period analyzed, with monthly visitors increasing from approximately 1,082 in January 2024 to 1,892 in April 2025.
Monthly Visits
Combined HTTP + HTTPS visits from January 2024 to June 2025.
Country-attributed Bandwidth
AWStats classified bandwidth activity by country or domain code. These classifications are useful for identifying broad patterns, but they should be interpreted cautiously because they can reflect ISP or cloud infrastructure rather than the visitor's actual physical location.
Search Behavior
Search activity provided another way to understand what visitors were looking for on the website. One of the clearest changes was the increase in searches related to the site's mining resources.
Share of recorded searches related to potentielminier.
Share of recorded searches related to potentielminier.
Interest in mining-related content increased significantly during the period analyzed. This was also reflected in the document-download data, where the mining-potential notice became the most downloaded tracked document.
Document Downloads
The website also provides visitors with access to downloadable documents. I analyzed the available download data to identify which types of documents were requested most frequently and to understand their potential impact on bandwidth usage.
Bandwidth Usage
Bandwidth was an important part of the analysis because the website had a usage limit. High traffic and frequent document downloads could increase bandwidth consumption and potentially create problems if the limit was reached.
Monthly Bandwidth Usage
Combined HTTP + HTTPS bandwidth usage compared with the 9.77 GB monthly limit.
Bandwidth by File Type
PDF files represented the largest identified file-type category in the AWStats breakdown.
Document Download Impact
Download frequency and bandwidth impact do not always tell the same story.
of the site's recorded bandwidth was associated with tracked document downloads during the analysis period.
These percentages measure overlapping aspects of the same traffic, so they should not be added together. For example, a bot downloading a PDF is counted as both automated traffic and document-download traffic.
of total recorded site bandwidth was associated with traffic identified as robots or automated clients.
Key Findings
- Website traffic showed significant overall growth during the period analyzed.
- The United States represented the largest share of page views.
- Haiti represented a significant portion of the audience and its share increased during 2025.
- Document downloads provided useful information about the resources visitors were most interested in.
- Bandwidth usage needed to be considered alongside traffic and document downloads because of the website's bandwidth limit.
Recommendations
- Prioritize optimization of frequently downloaded and bandwidth-heavy documents.
- Compress and reduce the size of large PDF files where possible.
- Track bandwidth, document downloads and traffic together to identify unusual increases in resource consumption.
- Review automated traffic and determine whether unnecessary crawling or repeated requests can be reduced without affecting useful search-engine indexing.
- Investigate website assets such as JavaScript and images, which also represented significant portions of bandwidth usage.
- Establish a simple monthly monitoring report comparing bandwidth usage against the 9.77 GB limit.
What I Learned
This project helped me understand how website data can be used to identify both opportunities and potential problems. I learned how to organize and analyze data in Google Sheets, compare different periods, identify visitor trends and interpret geographic and download data. I also learned that website traffic is not only about the number of visitors. The resources those visitors use can have a significant impact on the website, especially when bandwidth is limited.
Skills Demonstrated
Key Takeaways
- Website traffic increased significantly during the period analyzed.
- Bandwidth increased faster than visitor traffic and eventually reached the monthly limit.
- Document downloads were a major part of recorded bandwidth usage.
- Search activity and document downloads revealed increased interest in mining-related content.
- Automated traffic was another important factor to consider when analyzing bandwidth.
CASE STUDY
Receipt to Inventory & Analytics
Why I Built It
Receipts contain useful purchase data, but once they are stored away that information is difficult to search, compare or reuse. I wanted to explore a practical way to turn everyday receipts into structured data that could support household inventory and spending analysis while keeping the workflow on my own server.
Objective
The goal is to build a self-hosted workflow that can accept receipts from multiple sources, preserve the original documents, extract useful fields and line items, clean and organize the data, update inventory when appropriate, and make the structured information available for analysis.
The Challenge
Turning a receipt into useful data is more than an OCR problem. The workflow has to deal with inconsistent documents and protect downstream inventory and analytics from poor-quality data.
- Receipts arrive in different layouts, image qualities and file formats.
- OCR and AI extraction can return incomplete or inconsistent values.
- The same product can be described differently across stores and receipts.
- Duplicate receipts and low-confidence data need to be caught before they affect inventory or analytics.
- The process needs consistent business rules while keeping purchase data inside the self-hosted environment.
System Architecture
The design separates document handling, automation, local AI extraction, structured storage, inventory and analytics. PostgreSQL is intended to act as the central structured source of truth, with Grocy using validated inventory data and Metabase reading the database for analysis.
Planned end-to-end architecture for receipt ingestion, structured storage, inventory and analytics.
Data Flow
The workflow is designed as a sequence of small stages so that document storage, extraction, validation, inventory and analytics remain separate and easier to troubleshoot.
Ingest
Receive a mobile photo, uploaded PDF, forwarded email or manually added receipt.
Archive & OCR
Paperless-ngx preserves the original document and provides OCR text and metadata.
Optimize
Stirling-PDF can optionally standardize or improve the document before extraction.
Orchestrate
n8n watches for new documents, retrieves files, routes processing, handles retries and records workflow status.
Extract
Ollama with Qwen2.5-VL is designed to return store, date, totals and line items as structured JSON.
Validate & Normalize
Clean values, standardize products and categories, check confidence and detect duplicates before downstream use.
Store & Sync
PostgreSQL stores the structured records and inventory-eligible products can be synchronized with Grocy.
Analyze
Metabase can read the structured database to support spending trends, comparisons and custom dashboards.
Database Design
PostgreSQL is designed as the source of truth for structured data. The schema separates receipt transactions from reusable product information and processing history so the same validated data can support both inventory and analytics.
stores
Store information and store-specific rules.
receipts
One structured record per receipt, linked to its source document and store.
receipt_items
Line items from each receipt, including the original description and purchase values.
products
Canonical products used to normalize receipt descriptions and link eligible items to Grocy.
categories
Product categories and inventory-related classification.
brands
Reusable brand records linked to products.
processing_log
Workflow status, errors and processing history for each document.
Automation & Business Rules
n8n acts as the workflow orchestrator. The design keeps business rules explicit so extracted data can be checked and transformed before it reaches inventory or analytics.
Store Rules
Allow different tax, layout or parsing rules when stores format receipts differently.
Product Normalization
Map receipt descriptions to canonical products so the same item is not treated as a new product every time.
Category Rules
Assign categories using product information, store context and reusable rules.
Inventory Eligibility
Only send products intended for household stock tracking to Grocy.
Confidence & Duplicates
Flag low-confidence extractions and check Paperless/database identifiers to reduce duplicate processing.
Error Handling
Log failures and processing status so workflow problems can be reviewed and retried.
Data Quality
Extracting text is only the first step. The data needs to be cleaned, validated, standardized, categorized and checked for duplicates before it is reliable enough for inventory or analysis.
Keep the original OCR or extraction output as received so the source can always be reviewed.
Parse and clean the extracted values into a structured format before final validation.
Validate, standardize and categorize records so they are ready for operational use.
Aggregate and summarize validated data for dashboards and longer-term analysis.
Inventory
Grocy is intended to handle the operational side of the system. After validation and inventory-eligibility checks, suitable products can update household stock quantities, units and locations while keeping non-inventory purchases out of the inventory workflow.
Analytics
Metabase is intended to use PostgreSQL as a read-only analytics source. Because the project is still being developed, these are analytical capabilities rather than reported results.
Privacy & Self-Hosting
Receipts can contain personal purchase information, so the design keeps Paperless documents, local AI processing and PostgreSQL storage on the home server. Self-hosting gives me more control over where the data is processed and stored, although it does not make the system risk-free by itself.
Development Challenges
Current development work is focused on making the workflow dependable before relying on it for inventory or analysis.
- Improving OCR and image quality when receipt photos are difficult to read.
- Keeping AI output valid and consistently structured as JSON.
- Normalizing product descriptions that vary between stores.
- Detecting duplicates and defining confidence thresholds for manual review.
- Handling integration failures and retries without losing processing history.
What I’m Learning
Working on this project is helping me connect concepts that are often treated separately: unstructured documents, workflow automation, structured data, database design, validation, inventory and analytics. It is also giving me practical experience thinking about data quality before using data for decisions.
Skills & Concepts Demonstrated
Key Takeaways
- A useful receipt system needs data-quality rules, not only text extraction.
- A central PostgreSQL model separates document storage from reusable structured data.
- Inventory and analytics can share one validated dataset while serving different purposes.
- Local automation and AI processing keep the project aligned with the self-hosted infrastructure built in Project 01.