Milos N.
Senior AI/ML Engineer
Experience
Jan 2023 - Jul 2025
2 years 7 monthsSenior AI/ML Engineer
Meta AI
- Architected and implemented a production multi-agent system with a five-stage orchestration (Planner, Actioner, Executor, Feedback, Evaluator), achieving an 85% task success rate (+20% vs V2) through modular pipeline design and self-correcting feedback loops
- Built an end-to-end RAG retrieval pipeline with sliding-window chunking (50 lines + 10-line overlap), hybrid BM25+vector search, and parallel LLM summarization, improving retrieval precision by 85% and reducing hallucinations by 35%
- Developed Confluence and Jira integrations with full authentication, webhook support, and error handling to enable real-time knowledge base updates and cross-platform data synchronization for agent context
- Implemented an automated LLM-as-Judge evaluation framework with golden test cases, tournament scoring, and regression testing, shifting from subjective assessments to quantitative evaluation with 15+ benchmark configurations across 6 task categories
- Built TensorFlow/JAX/TFX pipelines on Vertex AI, Dataflow, and BigQuery for a web-scale ranking platform, improving CTR by 18% on 100M+ sessions/month and reducing p95 latency by 35% via feature-store redesign and hard negatives
- Launched privacy-preserving personalization using federated learning and differential privacy for PT/ES/EN markets, ensuring offline/online metric parity (AUC/PR, calibration) and automated drift alerts
- Standardized experimentation suite with A/B and interleaving tests, reusable metrics, and dashboards, reducing time-to-decision from 2–3 weeks to under 5 days
- Built a GenAI/RAG evaluation system with offline evaluator and guardrails, reducing hallucinations by ~35%, improving answer F1 by +7 points, and lowering p95 latency by 20%
- Ran Kubernetes/Docker microservices with model registry, shadow/canary deploys, and rollbacks; maintained a 99.9% inference SLO and MTTR under 10 minutes; mentored 6 data scientists and ML engineers and partnered with 4 product teams
Feb 2020 - Dec 2022
2 years 11 monthsData Scientist
Databricks
- Productionized SageMaker churn and propensity models with model registry, CI/CD, and blue-green deployment, reducing churn by 22% across three pilot cohorts (~45k users) with monitoring via MLflow and custom drift detectors
- Designed a real-time lakehouse data plane on S3, Glue, Athena, and EMR ingesting over 10 TB/day, and implemented streaming features with Kafka and Spark to enable ~1.8k QPS Lambda/Fargate inference
- Delivered LATAM regulated templates reference architectures that reduced time-to-production from ~3 weeks to 6 hours and lowered infrastructure costs by 18% through improved observability
- Implemented model governance with feature lineage, PII safeguards, and model calibration (ECE, Brier) to ensure consistent and auditable performance
Dec 2017 - Nov 2019
2 yearsPrincipal ML Consultant
Capgemini Invent
- Developed a Flask/React enterprise labeling platform to train and retrain CV/NLP models, reducing dataset turnaround time by 50% (4 weeks to 2 weeks) for a tier-1 bank and a public-sector client
- Deployed scikit-learn and PyTorch anti-spoofing and error monitoring models with a centralized Flask/DB2 error API, reducing critical incidents by 23% QoQ
- Built serverless identity management with Cloud Functions and Cloud SQL, lowering access-ticket resolution time by 30% and simplifying audits
Summary
- Senior AI/ML Engineer with over 10 years delivering production-grade AI that drives measurable business impact.
- Scope: Agentic AI, GenAI/RAG, ranking, NLP/CV, and large-scale experimentation; built calibrated, monitored, and drift-resilient ML systems (AUC/PR, ECE).
- Results: +18% CTR at web scale, -35% p95 latency, -22% churn, -18% infra cost.
Languages
English
AdvancedEducation
Aug 2013 - Jun 2015
Nanyang Technological University
Master of Science · Computer Science · Singapore
Aug 2009 - Jun 2013
Nanyang Technological University
Bachelor of Science · Computer Science · Singapore
Need a freelancer? Find your match in seconds.
Try FRATCH GPT More actions
Similar Freelancers
Discover other experts with similar qualifications and experience
- English
- Deutsch
2025 © FRATCH.IO GmbH. All rights reserved.