Cristian – Python, AWS, AI-assisted coding

Brand-protection platform on AWS EKS: search, advert, app-store and social-media detection pipelines.

• Owned full vertical slices from job enqueueing through Go consumers and headless crawlers to evidence capture, canaries, staged customer rollout, and legacy removal.
• Designed and shipped Microsoft Bing Ad Library as a production advert-search source with a KEDA-scaled Go consumer, screenshot evidence, customer rollout, and hardening for API limits and geo constraints.
• Built a standalone Go consumer for Brave's LLM Context API.
• Migrated generic and Samsung Galaxy Store app search onto headless crawlers and Go pipelines, increasing Galaxy Store results by about 5x.
• Cut customers over at 5 / 25 / 50 / 100 percent and then removed the legacy automation.
• Moved store enrichment for iTunes, Samsung, Huawei, and Baidu from Perl into Go.
• Migrated TikTok, YouTube, Weibo, and Pinterest social search to Go on Kubernetes.
• Added TikTok Ad Library ingestion and YouTube video evidence.
• Improved YouTube search to be about 6x faster with 2x more results.
• Improved Pinterest search to be about 3x faster.
• Added Spotify and Venmo screenshot evidence and canaries.
• Contributed to the WhatsApp reported-link ingest services.
• Rebuilt screenshot storage and retention with a high-availability screenshot API, logarithmic retention cleaner, content-hash deduplication groundwork, Prometheus metrics, and a Grafana dashboard.
• Built and operated the crawler canary and end-to-end reliability programme across 20+ platforms.
• Ran large lint and deprecation clean-ups across the Go monorepo.
• Designed 30+ guardrailed agent skills for Cursor.

Synthetic data-generation service inside an LLM evaluation platform, owned end to end as a member of the platform team.

• Designed and delivered the service that generated synthetic question/answer pairs and multi-turn conversations from customer documents for RAG experimentation and model assessment.
• Built asynchronous batches of up to 500 pairs on workers with transient-versus-terminal retry classification, idempotent submission with database deduplication, fair scheduling per API key, cancellation, and results stored as MLflow evaluation datasets.
• Put the generation engine behind a strict adapter boundary with a stub engine for tests.
• Added acceptance filters that rejected low-quality pairs.
• Enforced organization-scoped tenancy and attribute-based access control on every request.
• Worked inside the LLM-as-judge evaluation framework that scored MLflow traces on workers.
• Added pre-flight validation of model endpoints and document sources.
• Hardened the jobs API to return proper errors instead of 500s.
• Wrote architecture documentation covering service boundaries, data flow, worker topology, failure handling, and runbooks.
• Restructured build and delivery with layered Docker images, cutting incremental rebuilds from about 40 minutes to under 10.
• Split the data-generation service out of the platform's single pipeline into independently deployable units with path-filtered CI and per-service applications.
• Reduced pipeline execution by up to 80%.

Two fintech clients under NDA: a digital wealth-management platform and a small-business lending platform.

• Modernized projection, portfolio-simulation, and retirement-planning services across Go, Python, and Node.js in an Nx monorepo with Protobuf-defined APIs.
• Integrated ML forecasting modules.
• Worked on the Elasticsearch-backed search indexer.
• Added an LLM-backed explanation service in FastAPI that turned projection outputs into plain-language summaries, with structured JSON validated against the projection schema and guardrails against giving financial advice.
• Built partner marketplace services on Node.js and AWS serverless using Lambda, DynamoDB, EventBridge, SNS, and SQS.
• Built a PHP partner API with SQS-driven webhooks.
• Built a React and TypeScript submissions portal.
• Cut API response times by about 40%.
• Deployed 15+ containerized services to AWS EKS and Azure AKS with OpenTofu and GitHub Actions.
• Roughly halved CI/CD failures.

Genetic-testing company's processing platform and user-facing applications.

• Took operational ownership of the genetic-testing processing and user applications.
• Resolved recurring incidents and slow SQL.
• Fixed workflow bottlenecks and documented every fix, cutting recurring support workload by roughly 80%.
• Improved application performance by about 50%.
• Built Amazon QuickSight dashboards over RDS and CloudWatch data.

Health resource monitoring platform used in crisis-affected, low-connectivity regions.

• Redesigned parts of the architecture and database layer.
• Moved dashboard search and aggregation onto Elasticsearch, bringing dashboard load time from 20 seconds to 2 and raising supported country capacity by about 50%.
• Integrated Azure AD single sign-on for staff access and SOAP integrations with partner systems.
• Led the three-person team through architecture decisions, code review, and delivery planning.

Legacy English-learning platform with a web front end and native mobile apps.

• Cut application load times by about 40%.
• Migrated analytics from GA3 to GA4 through Google Tag Manager across 10+ tracking areas on web and mobile.
• Added OneTrust consent management for GDPR.
• Maintained and improved the containerized build and delivery pipeline.

Customer management platform for an energy company.

• Built a multi-threaded notification microservice for email and SMS with roughly 20x the previous throughput.
• Built code generators that replaced third-party providers, cutting messaging costs by about 70%.
• Reworked the multi-service authentication flow for a 5x speed-up.
• Added a thread-optimized audit system.

Modernization of a legacy cybersecurity (SIEM) platform.

• Owned architecture, development, and delivery while leading five developers on the modernization of a legacy product.
• Designed the new architecture and led the move off the legacy codebase.
• Delivered direct feature injection that removed about 80% of administrator setup.
• Integrated enterprise identity with Azure AD, ADCS / SCEP, PKI, and GPO.

VoIP applications for a telecom operator, built with a sole developer on the operational stream.

• Built the full user-management and operations platform covering user administration, billing, job processing, and reporting in one back-office application.
• Built an administrator panel with live charts.
• Implemented database-driven parallel job processing, making runs about 20x faster than the previous sequential approach.
• Automated client-specific billing, cutting manual work to roughly 10%.
• Built 10 operational KPI reports.