LM Lazar Milosavljević

Senior Software Engineer — Backend & Distributed Systems

Belgrade, RS

I build backend systems that have to stay up, and the products that sit on them.

Fifteen years of production software, most of it backend. The last five went to a data-intensive platform serving customers across sixteen countries: event-driven services on AWS, computer vision I trained and shipped myself, and a monthly cloud bill that fell while the platform grew.

ttfb
·····
transfer
·····
dom nodes
·····
frame
·····

Measured in your browser, right now

01 / Experience

Where the time
actually went.

  1. Apr 2021 — Present current Belgrade

    Data Genome

    Founder

    Independent software company, sole operator. Since June my attention has been on two products of my own.

    • Tamanas live

      A wishlist manager. The web application is live; iOS and Android are in review.

      tamanas.com
    • LAIN in development

      An AI nutrition assistant. In development, and not ready to be shown yet.

      datageno.me/lain
    • Java
    • Spring Boot
    • PostgreSQL
    • Kotlin
    • Python
    • Next.js / TypeScript
    • Flutter
    • Docker
    • Terraform / Terragrunt
    • AWS
  2. Apr 2021 — Jun 2026 Belgrade — remote, contracted through Data Genome

    Ourcart

    Senior Software Engineer, Backend

    Senior backend engineer on the core platform. Owned the design of major features and services, served as primary technical point of contact for the largest customer-facing product, and stood in for the VP of Engineering during absences.

    Platform ownership & delivery
    • Piloted the launch of a mobile application and its backend for a major market research company, taking it from three markets to sixteen — each country its own launch — and owning technical delivery end to end.
    • Led the backend and served as main technical point of contact for a market research platform used by around 100,000 users, covering architecture, delivery, and production support behind the internal support team.
    Cloud architecture & cost
    • The go-to engineer for the platform’s most complex AWS work, executed with no data loss and downtime minimized to the rare cases where it was unavoidable.
    • Helped drive the migration from a single legacy AWS account to a multi-account Control Tower architecture, cleanly separating production, staging and development environments and isolating products. With no existing infrastructure as code, built out the Terragrunt project covering every Ourcart product together with a colleague and an external AWS specialist partner, owning the application-specific and security-sensitive portions.
    • Reduced monthly cloud spend by over 30% across the Receipt Processing platform and Market Research product, while the number of moving parts and products grew. Levers included decommissioning unused resources, substituting cheaper equivalent services, instance-type changes and right-sizing, resource reconfiguration, and architectural redesign of services.
    • Cut S3 data transfer costs by consolidating non-us-east-1 buckets into us-east-1, configuring S3 Gateway Endpoints and updating services to route through them, alongside lossless compression of uploaded snapshots, expanded CloudFront usage and additional object caching.
    Reliability, security & incident response
    • First responder for production incidents and security events, coordinating response and communicating directly with the internal support team and other stakeholders.
    • Identified and shut down live attacks including a botnet exploiting the market research app’s referral system, coordinated bot activity from MTurk workers, a registration flood that was driving Twilio costs sharply upward, a firewall breach (resolved by reconfiguring WAF rules, which then blocked the large majority of malicious traffic), and bot attacks against a customer’s page farming reward points with fabricated receipts.
    • Built real-time monitoring of the platform’s components with automated alerting to the engineering and support teams, and wrote the internal postmortems following significant incidents.
    • Built internal malicious-activity detection and prevention, and integrated external fraud-signal tooling (IPQS).
    Applied AI & fraud detection
    • Cut daily human verification from ~3,000 receipts to under 500 by building the production visual anti-fraud system covering the platform’s receipt pipeline, running against 100,000+ receipt images per day. Most remaining human checks were confirmations of the system’s own conclusions.
    • Designed the system as a set of specialized detectors covering generator-produced and AI-generated receipts, handwritten and hand-modified receipts, digital and physical manipulation, screen captures and photos-of-screens, unauthorized fiscal-printer output, composite receipts assembled from multiple originals, and duplicate/series manipulation, all feeding a unified fraud scoring model that returned a final score with supporting metadata and human-readable reasons. Reached 98%+ fraud detection.
    • Owned the machine learning end to end: data preparation and augmentation strategy, training process design, model training, and the evaluation methodology used to select what shipped.
    • Proposed, designed and owned a product category-detection service, a capability I pitched internally and took from idea to production once a customer need emerged. Classified around 10,000 products per day into a custom taxonomy at roughly 99% accuracy, matching a GS1 reference dataset exactly on validation. Shipped on mostly existing infrastructure, adding effectively no incremental cloud cost while opening a new revenue line.
    Research
    • Designed and ran a multi-agent research program on the company’s hardest open problem. Defined specialized agent roles plus an adversarial reviewer role tasked with challenging every proposal for edge cases. The run produced 63 proofs of concept, ~15 of which held up and became production components, and yielded the cookbook and roadmap that six engineers then worked from on the Fraud Detection epic, after earlier in-house attempts had stalled.
    Technical leadership
    • Authored the design documents behind major changes. The team ran significant work as a research task producing a written design proposal, which then drove the implementation epic. I wrote these for the features and services I owned.
    • Onboarded several engineers onto the platform and its systems, and interviewed candidates for engineering roles.
    • Led the company’s adoption of AI-assisted engineering, tracking the field, piloting tools, and running internal evangelism and consultations.
    • Repeatedly drove architecture decisions to adoption, including designs selected over alternatives.
    • Java
    • Spring Boot
    • Python
    • TypeScript
    • AWS
    • Kinesis
    • Terraform / Terragrunt
    • CodeBuild / CodePipeline
    • MySQL
    • DynamoDB
    • Redshift
    • OpenSearch
    • Docker
  3. Oct 2014 — Nov 2020 Belgrade — part-time to Aug 2018, full-time thereafter

    Univerzitet Metropolitan

    R&D Software Engineer

    Research and development on the university’s e-learning platforms and institutional systems.

    • Built features across the learning management system used by the entire student body and teaching staff for course materials and assessment, and the authoring tool professors used to construct teaching content.
    • Rebuilt the full-text search over terabytes of course material, including rich media, on Apache Solr, taking query time from several minutes to milliseconds.
    • Designed and implemented an interactive mind map of each subject’s body of knowledge, letting students navigate to what they needed, see how topics connected, and identify their own gaps instead of reading sequentially or searching blindly.
    • Implemented cross-subject reuse of body-of-knowledge materials, eliminating duplicate authoring of shared content.
    • Built the university’s platform for professional short programs and courses.
    • Delivered lab instruction in C/C++ and rapid application development to visiting secondary-school groups and to adults in career-retraining programs.
    • Java
    • Spring
    • C#
    • ASP.NET
    • PHP
    • JavaScript
    • MySQL
    • Apache Solr
  4. Apr 2014 — Dec 2018 Belgrade — part-time, alongside Univerzitet Metropolitan

    Digi-Electronic d.o.o.

    Software Engineer

    Commercial products in the retail, accounting and fiscal-compliance domain.

    • digiPOS: POS and CRM platform used daily by 500+ companies in Serbia. Built the surrounding systems — a multi-branch synchronization service propagating internal documents with correct downstream accounting semantics (internal delivery notes triggering calculations and price levelling, price and price-type changes), and digiPOSmobile, an API and Android application letting operators place orders, check warehouse availability and track inventory.
    • digiTelRac: telecom expense management for corporate mobile contracts with Telenor (now Yettel), tracking usage and costs per line.
    • Fiscal cash register servicing software, built to Serbian fiscal certification rules.
    • C#
    • ASP.NET
    • Java
    • JavaScript
    • MySQL
    • Docker
    • AWS
    • Azure

Earlier

  • Nov 2015 — Mar 2017

    SKYloggy · Founder

    A skydiving jump-logging platform and community, with verification designed to prevent falsified jump records, on a GraphQL API and Kubernetes. Well received in testing; the company wound down before launch when the founding team split.

  • Jan 2010 — Apr 2014

    DesDev · Founder

    Web development studio. Ten-plus client projects, mostly websites and e-commerce stores, from requirements through deployment.

02 / About

The short
version.

I have been building software commercially since 2010, when I started a small web studio while I was still at school. For most of my degree I held two engineering jobs at once. Fifteen years on, the work has changed completely and the part I enjoy has not: designing a service, writing it, and looking after it once it is live.

Most of my week went on designing and writing software. Around that I was the primary technical point of contact for the largest customer-facing product, onboarded engineers, interviewed candidates, and led the company's adoption of AI-assisted engineering. When the VP of Engineering was away, I covered the role.

Significant work on my last team began as a written proposal: what needed to change for somebody, what we were giving up, what it would cost to run. That document was where the arguing happened, which was the point of it. I wrote them for the services I owned, and more often than not mine was the design that got built.

I pay attention to what software is for, not only whether it works. Product classification is the case I point at: I pitched it internally before any customer had asked for it, so when the need did appear the design was already argued out. It shipped on infrastructure we were already paying for and opened a new revenue line.

These days my attention is on my own products, and I am still reachable if the right role turns up.

03 / Capabilities

Where I'm useful.

Architecture & systems
Distributed systems, system design, event-driven, streaming and asynchronous architectures, microservices, REST API design, design patterns, scalability, performance and cost engineering
Cloud & infrastructure
AWS (EC2, Lambda, ECS, RDS/Aurora, DynamoDB, Redshift, S3, Kinesis, OpenSearch, ElastiCache, CloudFront, CloudWatch, WAF, IAM, SQS/SNS, SageMaker), multi-account architecture with AWS Control Tower, Terraform/Terragrunt, Docker, CI/CD, monitoring and alerting, cloud cost optimization
Data & AI
Data pipelines, ETL, data modeling, search at scale (Elasticsearch/OpenSearch), applied machine learning and computer vision, model training and evaluation, fraud detection systems, LLM integration, agentic AI workflows
Languages & frameworks
Java, Spring Boot, Python, TypeScript, JavaScript, React, Next.js, SQL, Bash
Databases
PostgreSQL, MySQL, DynamoDB, Redshift, MongoDB
Engineering leadership
Technical ownership, design documents and technical writing, incident response and postmortems, onboarding and mentoring engineers, interviewing, cross-functional coordination, stakeholder communication

04 / Education

On paper.

Languages

Native Serbian. Professional working proficiency in English (CEFR C1).

Open to the right role © 2026 Lazar Milosavljević © 2026 LM
offset 0/0