Rhesis AI GmbH: Backend & Infrastructure Engineer
Aug 2025 – Present · Germany
Infrastructure & Platform:
- Contributing to Rhesis, an open-source GenAI testing platform that enables cross-functional teams to design and automate quality validation for AI applications collaboratively.
- Provisioned all GCP infrastructure (GKE, networking, IAM, storage) using Terraform, deploying consistently across 3 isolated GCP projects.
- Implemented a GitOps delivery pipeline using ArgoCD with the App of Apps pattern, enabling fully automated deployments triggered end-to-end via GitHub Actions.
- Designed and maintained Helm-based Kubernetes deployment workflows for 10+ microservices across dev, staging, and production.
- Configured Kubernetes-native ingress routing using kGateway and HTTPRoute, and deployed WireGuard VPN with BIND9 as internal DNS for secure developer access.
- Deployed PostgreSQL using a tiered strategy — Helm subchart for dev, CloudNativePG Operator for staging/production — with automatic failover and continuous backup.
- Eliminated manual secrets/certificate management by integrating External Secrets Operator with GCP Secret Manager and cert-manager with Let’s Encrypt.
- Deployed a production observability stack (Prometheus, Loki, Alloy, Grafana) with custom OpenTelemetry instrumentation across all environments.
- Designed and deployed a custom containerized 8B-parameter LLM on Vertex AI using vLLM (8-bit quantization), reducing inference latency by 40%.
- Investigated and resolved Kubernetes production issues including pod crashes, container failures, and ingress routing problems.
- Diagnosed and resolved Linux runtime issues affecting containerized Kubernetes workloads.
Backend Development:
- Developed task management features in a FastAPI backend, including entity modeling, REST APIs, validation logic, and asynchronous notifications.
- Built a comment service that worked across multiple entity types letting users add comments directly to whichever one they were working with.
- Built and deployed Polyphemus, a Python-based LLM inference service on GCP, using vLLM with 8-bit quantization to cut inference latency by 40% with authentication, rate limiting, and batch processing built in.
- Benchmarked Cloud Run against three Vertex AI serving approaches (Hugging Face Service, custom container, and vLLM) for Polyphemus, running concurrency and throughput load tests across all four to compare production stability and inform the deployment platform decision.
- Designed an OpenTelemetry-based observability system in Python, including backend instrumentation and analytics pipelines with anonymized user/org identifiers, supporting both cloud and self- hosted deployments
Mitigant GmbH: Backend Engineer
Feb 2022 – Feb 2025 · Germany
- Designed and implemented a Cloud Attack Emulation SaaS platform offering 50+ cloud attack scenarios to assess and improve customers’ AWS and Azure cloud security posture, based on the MITRE ATT&CK and ATLAS frameworks.
- Built backend microservices in Java (Spring Boot) using Infrastructure as Code (IaC) to provision and securely access AWS and Azure environments.
- Architected a RabbitMQ-based messaging backbone to decouple 8 microservices, enabling asynchronous, event-driven communication and supporting zero-downtime multi-cloud deployments.
- Took full ownership of the Azure cloud attack emulation feature from design to deployment and monitoring, helping attract multi-cloud enterprise customers to the platform.
- Integrated Amazon Bedrock APIs to implement GenAI-based attack simulations, strengthening AI security evaluation in production systems.
- Integrated AWS CloudTrail, Amazon GuardDuty, and Datadog to enable real-time monitoring, observability, and evidence collection during simulated cloud attacks.
- Implemented comprehensive unit and integration tests using JUnit and Mockito, achieving over 80% code coverage and improving system reliability.
- Used Redis for caching and real-time user status tracking to deliver simulation reports to customers when online.
- Collaborated with the frontend team to design APIs supporting dynamic configuration of attack scenarios and real-time feedback during simulations.
- Conducted peer code reviews enforcing clean architecture principles, reducing integration issues and improving production deployment stability.
Hasso Plattner Institute: Student Assistant
Jul 2021 – Feb 2022 · Germany
- Designed and developed the frontend and backend of a web application to process and visualize data.
- Processed and modeled data using Java and Gson, storing it in PostgreSQL for efficient querying and backend performance.
- Developed an interactive React/JavaScript frontend with advanced filtering and pagination, reducing load time by 50%.
- Implemented backend caching strategies using Redis to improve response times for frequently accessed data, resulting in a 40% reduction in database load.
BRAC University: Full-stack Engineer
Dec 2017 – May 2021 · Bangladesh
- Developed new features for the University Student Information System — including online admission, student advising, accounts, and grade submission — supporting over 20,000 students.
- Optimized backend performance using HikariCP connection pool tuning, database indexing, and Redis caching to handle peak system load, improving response times by 30%.
- Designed and implemented a role-based access control (RBAC) system to manage user permissions and enhance security across the application.
- Implemented an audit trail system using Spring Security to track access and changes to sensitive data such as student payments and grades.
- Integrated payments via the SSLCommerz payment gateway, reducing manual processing errors by 30%.
- Built responsive JSP/HTML/CSS interfaces integrated with Groovy/Grails backend services for scalable web applications.
CS InfoTech Ltd.: Full-stack Engineer
Dec 2016 – Dec 2017 · Bangladesh
- Delivered full-stack features for an automated loan-management application using Java, Spring Boot, Hibernate (JPA), and MySQL on the backend, and Thymeleaf on the frontend.
- Designed and implemented a credit scoring module that processed over 1,000 applications per month, improving risk assessment consistency and decision-making speed.
- Integrated the LogiDoc document management system, cutting document retrieval time by 60% and digitizing workflows.
- Implemented a notification system using Spring Boot and JavaScript to automate loan status updates, improving customer communication and reducing manual follow-ups.
- Applied Test-Driven Development (TDD) and wrote unit tests using JUnit to ensure correctness and maintainability of backend modules.