AI/ML Engineer
Cyber Security Umbrella · Remote
2024.12 – 2025.09 · 10 mos
Deployed GenAI and compliance systems for cybersecurity operations, including RAG assistants and fine-tuned LLM workflows. I mainly work with RAG, FastAPI, LangChain, AWS to ship useful AI products and workflow automation.
AI/ML Engineer at Cyber Security Umbrella focused on shipping production systems with end-to-end ownership.
I handled solution design, implementation, and delivery while collaborating with cross-functional teams to align engineering execution with business goals.
- Scope: Deployed GenAI and compliance systems for cybersecurity operations, including RAG assistants and fine-tuned LLM workflows.
- Ownership: architecture, implementation, and rollout
- Context: high-impact business workflows and reliability requirements
- Business pain point addressed by: Deployed GenAI real-time assistant handling 82,000+ cybersecurity scenarios using RAG + LangChain + Gemini API — 40% faster incident triage.
- Business pain point addressed by: Built multi-model compliance system using LoRA/QLoRA fine-tuned LLMs achieving ~95% operational accuracy in regulatory mapping.
- Business pain point addressed by: Designed SOC analytics pipeline aggregating data from 6+ security tools with real-time ingestion and anomaly detection.
- Business pain point addressed by: Led cross-functional team of 5 engineers using agile ML workflows, accelerating delivery by 25%.
Project 1
- Business problem: Manual or slow workflow was limiting business throughput.
- My contribution: Deployed GenAI real-time assistant handling 82,000+ cybersecurity scenarios using RAG + LangChain + Gemini API — 40% faster incident triage.
- Technologies: Python · FastAPI · LangChain · AWS SageMaker · LoRA/QLoRA
- Outcome: Delivered measurable reliability and speed improvements.
Project 2
- Business problem: Manual or slow workflow was limiting business throughput.
- My contribution: Built multi-model compliance system using LoRA/QLoRA fine-tuned LLMs achieving ~95% operational accuracy in regulatory mapping.
- Technologies: Python · FastAPI · LangChain · AWS SageMaker · LoRA/QLoRA
- Outcome: Delivered measurable reliability and speed improvements.
Automation Pipeline
- Business problem: Manual or slow workflow was limiting business throughput.
- My contribution: Designed SOC analytics pipeline aggregating data from 6+ security tools with real-time ingestion and anomaly detection.
- Technologies: Python · FastAPI · LangChain · AWS SageMaker · LoRA/QLoRA
- Outcome: Delivered measurable reliability and speed improvements.
Project 4
- Business problem: Manual or slow workflow was limiting business throughput.
- My contribution: Led cross-functional team of 5 engineers using agile ML workflows, accelerating delivery by 25%.
- Technologies: Python · FastAPI · LangChain · AWS SageMaker · LoRA/QLoRA
- Outcome: Delivered measurable reliability and speed improvements.
- Designed SOC analytics pipeline aggregating data from 6+ security tools with real-time ingestion and anomaly detection.
- Led cross-functional team of 5 engineers using agile ML workflows, accelerating delivery by 25%.
- Deployed GenAI real-time assistant handling 82,000+ cybersecurity scenarios using RAG + LangChain + Gemini API — 40% faster incident triage.
- Built multi-model compliance system using LoRA/QLoRA fine-tuned LLMs achieving ~95% operational accuracy in regulatory mapping.
- Designed SOC analytics pipeline aggregating data from 6+ security tools with real-time ingestion and anomaly detection.
- Deployed scalable solution on AWS SageMaker with auto-scaling FastAPI endpoints handling 1,000+ concurrent requests.
Challenge
- Deployed GenAI real-time assistant handling 82,000+ cybersecurity scenarios using RAG + LangChain + Gemini API — 40% faster incident triage.
Constraints
- Production reliability and latency expectations
- Cross-team coordination and evolving requirements
Solution
- Designed pragmatic architecture tradeoffs to balance quality, speed, and maintainability.
Outcome
- Shipped stable systems with measurable operational improvements.
- Used retrieval-augmented workflows when explainability and updatable knowledge mattered more than model re-training.
- Preferred relational storage for correctness, querying flexibility, and operational familiarity.
- Adopted asynchronous/concurrent processing where throughput and responsiveness were critical.
- Used hybrid extraction approaches to improve reliability across semi-structured inputs.
- Business impact comes from system reliability and usability, not model novelty alone.
- Clear ownership boundaries speed up delivery in multi-system initiatives.
- Incremental architecture decisions compound into maintainable platforms.
- Deployed GenAI real-time assistant handling 82,000+ cybersecurity scenarios using RAG + LangChain + Gemini API — 40% faster incident triage.
- Built multi-model compliance system using LoRA/QLoRA fine-tuned LLMs achieving ~95% operational accuracy in regulatory mapping.
- Designed SOC analytics pipeline aggregating data from 6+ security tools with real-time ingestion and anomaly detection.
- Led cross-functional team of 5 engineers using agile ML workflows, accelerating delivery by 25%.
- Deployed scalable solution on AWS SageMaker with auto-scaling FastAPI endpoints handling 1,000+ concurrent requests.
AI
- LangChain
- LoRA/QLoRA
- RAG
- Gemini API
Backend
- Python
- FastAPI
- REST API
Cloud
- AWS SageMaker
Infrastructure
- Git
- Docker
- Monitoring
- Logging
- Concurrent Processing
Tools
- Anomaly Detection
- Agile