Selected work

Details

A home cooking assistant built for use on a phone in the kitchen. You tick the ingredients you have, then choose the meal type, cuisine, diet, calorie limit and kitchen equipment. Claude AI then suggests real, well-known dishes instead of made-up combinations. If you allow it, it can swap in a sensible substitute from your pantry. You can keep chatting about the suggestions, and the conversation picks up where you left off on any device.

To prevent made-up content, every dish story, chef tip or fun fact must quote a web page word for word. The app opens that page, checks that the quote is really there, and drops anything it can't confirm.

You can save favourite recipes to a personal cookbook. Each one is automatically filed under one of more than 30 cuisine families grouped by region, tagged by difficulty, cooking time and nutrition profile, and given a photo. Your star ratings shape future suggestions. Earlier recipes are reused or adapted instead of being generated from scratch every time.

The app runs on the home network and connects to an existing Claude subscription from the browser, so there are no extra AI costs. The private access key is never shown in the browser, the number of AI sessions running at once is limited, and the data is backed up automatically every day. More than 100 automated tests check that it works correctly.

• python
• sqlite
• docker
• fastapi
• tailwind css
• websockets
• anthropic claude
• alpine.js
• claude agent sdk

Details

CustomsWait tracks how long drivers really wait at road border crossings in South-East Europe. Travellers report their crossing time, GPS-verified at the border, and the site turns those reports into live waits, busy-hour heatmaps and best-time-to-travel advice across a home list, an interactive map, per-country pages and a detail page per crossing with live border cameras. Visitors can save favourite crossings and get email or push alerts when a queue clears. An admin back office runs crossings, moderation, a news blog, newsletters and visitor analytics. Everything is available in 11 languages, with automated scrapers filling the gaps between reports.

• python
• postgresql
• ubuntu
• docker
• fastapi
• javascript
• openstreet maps
• hetzner
• google analytics
• leaflet
• progressive web application

Details

Back-office platform with five sections.

The Billing module mirrors the team's ClickUp workspace every minute: progress against the monthly hour target, per-client totals by billing stage priced at each client's rate, PDF reports.

The Companies module loads the Romanian public company register, computes ~30 financial indicators and risk flags, and serves them over a token-authenticated API.

The Research agent is an external, Claude-based agent that researches the web for companies that could represent potential clients, taking into consideration a set of predefined rules and instructions, extracts contacts, analyses financial and litigation records, scores fit and drafts cold emails and saves everything to the main app via an API endpoint. It also interacts with the Companies module, from which it extracts financial and identification data for the companies it researches.

The Prospects module receives leads from the research agent via an API endpoint and stores them; leads are de-duplicated, tracked through a sales pipeline and printed as PDF cards. It serves as the central panel for all lead-related data. The sales personnel will check every lead data and confirm them.

The Outreach module is integrated with a company mailbox and, for each lead confirmed by the sales team, it will craft a profesional and spam-proof email and send it to the prospect, then it will monitor the mailbox for responses from the customer. On response it will send notifications to the person(s) responsible for the lead.

The Projects module ia a fully interactive project management suite that allows client and project tracking at task level and is tightly integrated with the Billing module. It also supports project and client level wikis, file attachments and comments.

• python
• pandas
• postgresql
• django
• django rest framework
• datatables
• docker
• hetzner
• anthropic claude
• clickup api

Details

Sole maintainer, since April 2025, of a Romanian B2B product catalog and partner portal. Modernized the deployment with a Dockerized environment, introduced a granular user-tier system controlling image download, text copy, and technical-documentation access, and overhauled the partner offer-request workflow with multi-recipient email notifications. Built a lockable PDF generator for product comparative tables, with localized headers/footers and QR-code lifecycle handling. Added category duplication, an emoji picker with UTF-8 SEO field migration, article character counters in admin forms, and rebuilt the accessories and similar-products carousels. Also fixed slug generation.

• php
• mysql
• jquery

Details

Web platform serving Romanian Christian communities with articles, news, an online Bible, a church directory, and a donations system. Since early 2025 I led a major overhaul: rebuilt the donations module and integration with various payment processors, created an admin-managed donation methods system; redesigned the homepage, donation cases, live churches, the search page, news section; implemented configurable SEO fields, sitemap and also applied page-speed optimizations, CSP security policies, GDPR cookie consent with server-side logging, a custom versioned query cache, redesigned the respurce import flow and created a resource import/deduplication pipeline.

• paypal
• tailwind css
• laravel
• php
• euplatesc

Details

Autonomous SQL query performance monitoring system for Azure DB. Continously scans Azure DB instances, extracts non-performant SQL queries and stored procedures and attempts to reformat them with the help of and OpenAI-based agent, in order to obtain a lower run time and lower compute resource consumption. It is aimed to generate savings by lowering cloud costs.

• python
• azure
• openai

Details

A Python application designed to safeguard website health by automatically detecting broken or unreachable links at scale. The tool recursively crawls a website's sitemap hierarchy - including nested sitemap indexes — and verifies every page URL via concurrent HTTP requests, applying retry logic with progressive timeouts to minimize false positives. Results are exported into structured CSV reports that flag dead pages, inaccessible sitemaps, and the full crawl path for each issue, giving site owners and SEO teams an actionable map of where to focus repairs. An interruptible, resumable workflow makes it practical to audit sites with tens of thousands of URLs.

• python

Details

A Python-based automation tool that provisions a private, self-hosted file server on cloud infrastructure with a single command. The application orchestrates the entire setup process: it creates a virtual server, configures secure remote access, provisions scalable object storage for file data, hardens the operating system against common threats, sets up DNS records and a reverse proxy with HTTPS, and deploys a fully functional Nextcloud instance for browser- and desktop-based file sharing. It tracks every resource it creates, supports clean teardown, and can resume safely after interruptions — turning hours of manual server setup into a repeatable, reliable pipeline.

• python
• nextcloud
• hetzner

Details

DSig is a multi-tenant SaaS platform for remotely managing fleets of digital signage displays running on remote edge devices. The cloud portal lets operators upload media, build playlists, schedule campaigns and monitor device health, while a resilient on-device player runs 24/7 and survives network outages by caching content locally. A real-time messaging layer pushes commands and live status between cloud and field devices, and a zero-touch provisioning flow turns a freshly-imaged unit into a managed display with a single setup command. Components include the management portal, async task workers, message broker, observability stack and the edge player.

• python
• postgresql
• flask
• django
• docker
• redis
• celery
• mqtt
• websockets
• s3
• grafana

Details

Web application that helps users discover points of interest around any chosen location, filtered by category and reachable within a set distance or driving-time budget. The interface centers on an interactive map that progressively renders results as they're found, keeping the experience responsive even with large result sets. Built on Django with a containerized deployment, the app supports both Google Maps and OpenStreetMap as interchangeable map providers, switchable through a single configuration setting. It also runs as a Progressive Web App, so users can install it on their device and continue using it offline.

• python
• google cloud platform
• postgresql
• django
• nginx
• ubuntu
• javascript
• openstreet maps

Details

Backend: API interacting with locally deployed Llama 3.1 8b LLM, through Ollama. It manages, not only interaction with the LLM, but also prompt creation and documentation pre-processing. External documentation is fed to the backend via uploaded files or API integration.
Frontend: Simple web interface built with HTML with Tailwind CSS and JavaScript.

• python
• docker
• fastapi
• ollama
• llama
• tailwind css
• javascript

Details

Built in Python, with Pandas and NumPy, fully dockerized, using a proprietary library for implementation DAG data processing approach.

• python
• pandas
• numpy
• docker
• redis
• rabbitmq
• celery
• azure

Details

Web application developed for the local branch of a major European banking group. It was used to keep track of the renewal timelines for tech supplier contracts. Developed using FastAPI + SQLite on the backend and Vue.js+Bootstrap on the frontend. Since the application was supposed to be multi-user, but, due to some security constraints, a database server was not available, I also had to devise a way to use SQLite over the network, while also allowing multiple users, from various locations, to interact and read/write from/to the database.

• python
• pandas
• sqlite
• bootstrap
• windows
• datatables
• server
• fastapi

Details

In-location company server running Nextcloud, on Ubuntu 22.04 LTS. Installed repurposed hard drive, extracted from WD Cloud NAS, deployed OS and Nextcloud service. Secured application with SSL certificate and server with firewall, configured port forwarding in router, and added necessary DNS records for the domain.

• ubuntu
• nextcloud

Details

Django application that serves as a presentation website, developer profile, and resume manager, with lead generator capabilities.

• python
• google cloud platform
• sqlite
• django
• nginx
• ubuntu
• xhtml2pdf

Details

Web application that serves as an e-commerce extension for the main WordPress-based website of the company. Functionalities: user management with role segregation, service listing, purchase & refund (integration with PayPal), order management, ticketing, internal notification system, emailing (integration with SendGrid), bulk data upload from CSV files, analytics & reporting, database template caching. The app is developed with Django, will run on DigitalOcean AppPlatform, and will use a DigitalOcean PostgreSQL managed database. A separate module manages the integration with various APIs in order to update the various metrics of the products/services that the company sells.

• python
• pandas
• sqlite
• postgresql
• bootstrap
• digitalocean
• django
• nginx
• ubuntu
• datatables
• vue.js
• paypal
• sendgrid

Details

NFC-enabled business cards service, backed by a custom-developed Django-based web service.

• python
• pandas
• postgresql
• django
• nginx
• ubuntu
• oracle cloud
• bulma
• nfc
• django rest framework

Details

An app that periodically scans Facebook groups for posts and extracts the ones deemed relevant (contain certain keywords), together with the contact information of the post author, and maintains a dataset with those leads and their status. The leads are then presented in human-readable form by a web app running on the local network.

• python
• pandas
• sqlite
• bootstrap
• django
• nginx
• ubuntu
• datatables

Details

Terraform project that deploys e-commerce applications, such as WordPress + WooCommerce, inside a scalable architecture on Digitalocean, using an auto-scalable Kubernetes cluster with a load balancer and volumes for the persistence of the application and database files, all inside a VPC created purposefully for the deployment.

• terraform
• wordpress
• woocommerce
• digitalocean

Details

An application that looks at all participants at events organized by the company and creates a list of medics that represent further commercial interest. The list is then enriched with relevant information collected from various systems. It does so by connecting via REST to Salesforce, and other systems for the extraction of information. This software is strongly integrated with the Windows OS, making use of its API to generate PDF files from the output files.

• python
• pandas
• salesforce
• excel
• windows

Details

In the world of pharmaceutical sales, the representatives of the company are each assigned to a territory inside which they carry out their activity. Given that the variables in this world are ever-changing (medics move away, reps resign, etc.) the company finds it hard to keep track and always have a clear image of what is going on with their sales force. The application extracts information from the client's information systems (via REST API) and updates the assignment of the reps to their respective territories. It is comprised of a back-end that provides the main functionality and a front-end that serves as a GUI and provides access to settings, logs, and statistics.

• python
• pandas
• google cloud platform
• sqlite
• gmail
• selenium
• flask
• bootstrap
• salesforce

Details

A cloud-native, multi-instance web application that collects hotel room prices that meet certain criteria and displays them, in an aggregated form, on the app website. It uses two AWS EC2 instances for the collector apps, an AWS S3 bucket for storage, and two Google Compute Engine instances for the front-end, database and maintenance/tasks.

• python
• pandas
• google cloud platform
• sqlite
• aws s3
• aws ec2
• selenium
• postgresql
• flask
• nginx
• ubuntu
• datatables

Details

Created a series of web scrapers that extract hotel information from various aggregators/hotel operators' websites.

• python
• google cloud platform
• aws s3
• aws ec2
• selenium

Details

An application that helps notary offices to quickly compute the fees for their services, in accordance with the guidelines of The National Union of Notaries Public of Romania, and also provides a detailed view of the various cost components.

• excel vba

Details

The application periodically accesses a web-based database to extract reports containing enrollment data for various company-owned websites. Due to the fact that there is no API access, this operation is performed via the front-end, with Selenium. The downloaded data is stored and then checked against the preexisting information, recorded in the company's Veeva instance, which is accessed via REST API. Because the enrollment data is filled in manually, the algorithm uses string matching techniques to overcome potential human error (typos, missing letters, special characters, etc.). The users that are already registered, are validated against a series of business rules provided by the client and divided by the system of provenience. The result is then output to files that are each sent to the person responsible for the respective system.

• python
• pandas
• google cloud platform
• sqlite
• gmail
• selenium
• salesforce

Details

Refactoring in Python of the Spend&Savings reporting tool, previously realized in VBA. The most laborious part of the process was reimplemented using the KNN machine learning technique for classification, based on categorical variables. Further reduced report realization time from a few hours to about 30 minutes.

• python
• pandas
• scikit learn
• windows

Details

The application takes in structured datasets that have been filled in manually and performs deduplication of records based on one or more fields. Given that the data sets were created manually and thus are prone to contain errors, the deduplication algorithm uses a series of score-based string-matching techniques to aid in the process.

• python
• pandas
• windows

Details

The application centralizes the data collected from the reports submitted by the field reps of the company. It then structures them in tabular form to make them ready for analysis and interpretation.

• excel vba

Details

Developed tools for automation of reporting for travel expenses, cell phone usage, monthly department activity, automated procurement bid forms, and data collection inside the department, for Procurement. Refactored and optimized for increased performance the Regional Business Plan for Sales, and also created Jira Ticket Report Updater for QA.

• excel vba

Details

Excel VBA-based application that takes multiple inputs and generates the monthly spend and savings report for use in the procurement department. By automating tasks it managed to drastically reduce the time for the realization of the report from days to hours.

• excel vba