Alexandru – Python, React, AWS

Enterprise-grade, low-code conversational AI platform that allows businesses to build and deploy intelligent chatbots and voice assistants that connect seamlessly with backend systems (like ERP, CRM, and SaaS applications) to automate routine tasks.

• Migrated training and inference Python code-base to newer library versions (TensorFlow, Keras, PyTorch, Transformers);
• Implemented a continuous delivery pipeline that reduced the deployment time by 40%;
• Troubleshot complex workflows involving multiple ML models through Python and Java code bases in a Kubernetes deployment;
• Designed and implemented sandboxes for ML models, local training, and inference using Docker, Python, and Bash scripting, increasing the error detection delay from 3h to 15min;
• Secured artifacts access by integrating secret manager services in existing CI pipelines and automated secret rotation using JFrog API.

Specialized AI tool for automated personal branding that leverages LLMs to generate high-engagement LinkedIn posts in a user’s unique voice.

• Implemented the business flows by leveraging the Workflow-as-a-Service cloud model, using TypeScript, Node.js, and Upstash;
• Created UI components using React, Next.js, and Tailwind with reactive updates using Convex;
• Secured the application with SSO login using Clerk;
• Streamlined the CI/CD process using GitLab CI and Vercel API;
• Built agent workflows by integrating the OpenAI and Claude APIs;
• Implemented multi-tier subscriptions using the Paddle API.

Enterprise ETL system for processing millions of Electronic Health Records (EHR) produced by multiple healthcare organizations

• Implemented transformations in Java 8 using Apache Crunch and MapReduce;
• Introduced common utilities leveraging Java 8 functional programming features to reduce the transformation implementation code by up to 60%;
• Created Python scripts for migrating from Snowflake to Oracle Autonomous Data Warehouse, using AWS S3 and Paquet;
• Coordinated technical discussions and initiatives in a team of five engineers.

A scalable, highly available API for extracting valuable insights from live data feeds (e.g., radio broadcasts, TV shows) by performing transcription, speaker detection, sentiment analysis, and summarization.

• Developed an event-driven micro-services system using Python and Java around the Orchestrator pattern;
• Created a Python framework isolating the reusable code of a stream processor, allowing the team to focus on the business logic'
• Configured a Pulsar cluster for handling large volume data flows between application;
• Implemented a centralized AVRO schema repository for domain model standardization and applied AVRO SerDe for efficient data transfers, achieving 4x less bandwidth consumption compared to JSON;
• Secured the system using the OIDC authorization protocol with a self-hosted Keycloak deployment;
• Dockerized all components and deployed as a Kubernetes cluster, leveraging the AWS EKS;
• Optimized resource utilization by implementing custom auto-scaling rules, using Keda and Karpenter;
• Implemented monitoring systems for collecting application metrics and logs using Prometheus Node Exporter, Promtail, Loki, and Grafana.

A cloud infrastructure modernization project focused on transforming manually managed environments into a scalable Infrastructure-as-Code (IaC) platform. The solution was designed to support customizable deployments across multiple tenants and regions, enabling consistent provisioning, simplified operations, improved maintainability, and faster delivery of infrastructure changes.

• Isolated system components into reusable Terraform modules and imported existing infrastructure;
• Templated infrastructure manifests using Terragrunt, splitting the Terraform centralized state into isolated states;
• Organized system components into Helm charts and grouped tightly-coupled components under Helmfiles;
• Integrated SOPS for encrypting secrets in version control and using them inside Terraform and Helm modules.

Web application for transcoding media content in real-time that enables the integration between a transcription system and live data feeds (e.g. microphone feed in browser, IPTV camera stream)

• Used Python multiprocessing primitives for handling parallel transcoding processes;
• Implemented inter-process communication using Unix socket file descriptors;
• Handled multiple connections using the Selector API for I/O multiplexing;
• Integrated Pulsar consumer and producer API for I/O transfer.

Implemented the end-to-end process for delivering the best-in-market STT model for the Romanian language, from data collection to model deployment.

• Developed the data annotation platform using Java, React, MongoDB, and GCS;
• Trained and evaluated a wav2vec model using Python, PyTorch, and Fairseq, managing to achieve a WER (Word Error Rate) down to 5%;
• Implemented a consumer-producer event-driven system for serving the model, using Python and RabbitMQ;
• Exposed an HTTP API using Flask.

An event-driven system that integrates with railway hardware and centralizes diagnostic data on a single platform, enabling efficient monitoring and alerting.

• Developed and deployed Spring, Java 11 micro-services to a self-managed Kubernetes cluster;
• Created and tested RESTful endpoints using Spring and RabbitMQ;
• Implemented persistence layer using JPA, Hibernate, PostgreSQL, and Liquibase;
• Performed testing with JUnit, Mockito, and Spring Mock MVC;
• Implemented a custom Keycloak plugin.

A healthcare imaging platform designed to securely retrieve, view, store, and share medical imaging data across multiple locations. The solution included a cloud-based PACS architecture and a zero-footprint DICOM viewer, enabling healthcare professionals to access and collaborate on diagnostic images without requiring local software installation, while maintaining security, scalability, and regulatory compliance.

• Developed and tested RESTful web services with Spring, Java 8, and implemented Reporting Database pattern using Apache Solr;
• Implemented persistence layer using JPA, Hibernate, PostgreSQL, and Liquibase;
• Load tested the web platform by simulating various user flows using JMeter, BlazeMeter, and Selenium.

Web platform for processing multi-media files, including transcription, translation, and AI-enabled insights.

• Created backend and frontend applications using React and Java Spring;
• Leveraged AWS EKS and Docker for deployment, and Cloudflare R2 for file storage;
• Integrated Stripe API for both subscriptions and one-time payments;
• Implemented upload flow in AWS Step Functions for handling large uploads.