Looks like you have JavaScript disabled. For the full Twine experience, you will need to re-enable it.

Agentic AI Safety Architect and Staff Systems Engineer with 25+ years of production experience building and deploying real-world agentic AI systems across AWS, GCP, and Azure. I design and orchestrate autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots with production-grade orchestration and strict human-in-the-loop controls. I architect full-stack agent capabilities (perception, planning, action), implement rigorous LLMOps evaluation pipelines, and enforce ethics/safety guardrails through contract boundaries, memory/state management, and sandboxing layers. I also fine-tune and self-host open-source models on private GPU clusters, with a focus on preventing unsafe tool access and blocking insecure code from reaching production.…Agentic AI Safety Architect and Staff Systems Engineer with 25+ years of production experience building and deploying real-world agentic AI systems across AWS, GCP, and Azure. I design and orchestrate autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots with production-grade orchestration and strict human-in-the-loop controls. I architect full-stack agent capabilities (perception, planning, action), implement rigorous LLMOps evaluation pipelines, and enforce ethics/safety guardrails through contract boundaries, memory/state management, and sandboxing layers. I also fine-tune and self-host open-source models on private GPU clusters, with a focus on preventing unsafe tool access and blocking insecure code from reaching production.

Serge Shuster

AI Developer, AI Ethics Consultant, AI Evaluator, +2





Agentic AI Safety Architect and Staff Systems Engineer with 25+ years of production experience building and deploying real-world agentic AI systems across AWS, GCP, and Azure. I design and orchestrate autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots with production-grade orchestration and strict human-in-the-loop controls. I architect full-stack agent capabilities (perception, planning, action), implement rigorous LLMOps evaluation pipelines, and enforce ethics/safety guardrails through contract boundaries, memory/state management, and sandboxing layers. I also fine-tune and self-host open-source models on private GPU clusters, with a focus on preventing unsafe tool access and blocking insecure code from reaching production.…Agentic AI Safety Architect and Staff Systems Engineer with 25+ years of production experience building and deploying real-world agentic AI systems across AWS, GCP, and Azure. I design and orchestrate autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots with production-grade orchestration and strict human-in-the-loop controls. I architect full-stack agent capabilities (perception, planning, action), implement rigorous LLMOps evaluation pipelines, and enforce ethics/safety guardrails through contract boundaries, memory/state management, and sandboxing layers. I also fine-tune and self-host open-source models on private GPU clusters, with a focus on preventing unsafe tool access and blocking insecure code from reaching production.

Available to hire

Agentic AI Safety Architect and Staff Systems Engineer with 25+ years of production experience building and deploying real-world agentic AI systems across AWS, GCP, and Azure. I design and orchestrate autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots with production-grade orchestration and strict human-in-the-loop controls.

I architect full-stack agent capabilities (perception, planning, action), implement rigorous LLMOps evaluation pipelines, and enforce ethics/safety guardrails through contract boundaries, memory/state management, and sandboxing layers. I also fine-tune and self-host open-source models on private GPU clusters, with a focus on preventing unsafe tool access and blocking insecure code from reaching production.

Skills

Experience Level

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Expert

Intermediate

Beginner

Beginner

Language

English

Fluent

Work Experience

Agentic AI Safety Architect & Multi-Agent Systems Auditor at Magical Industries

January 1, 2024 - Present

Architect and orchestrate agentic deployments including autonomous coding assistants, multi-agent research teams, RAG knowledge agents, and customer-support bots using LangChain/LangGraph, CrewAI, AutoGen, LlamaIndex, Semantic Kernel, and PydanticAI across AWS/GCP/Azure. Use MCP and A2A to define secure tool use and cross-agent communication with contract boundaries. Deploy self-hosted LLM-driven coding agents on private GPU clusters (vLLM/Ollama/SGLang) with human-in-the-loop approval gates for high-risk actions. Build multi-agent workflows (e.g., deterministic research/report generation) with Temporal orchestration and rollback on failure. Apply LoRA/QLoRA fine-tuning using Unsloth/Transformers and implement ethics/safety guardrails for production release, including memory/state handling to prevent sensitive-data leakage and unsafe execution persistence.

Systems Engineer & Agentic Code Safety Lead at Wells Fargo

January 1, 2020 - January 1, 2024

Established enterprise-wide code-review and static-analysis pipelines for C++, Java, and Python, enforcing safe concurrency patterns and memory-safety practices. Led production incident response for low-latency C++ systems, addressing memory leaks and race conditions and improving outage reduction through root-cause debugging. Built and hardened real-time analytics engines handling millions of transactions daily, with focus on finding undefined behavior under load. Implemented Python/C++ interop with strict validation gates preventing unsafe memory access from AI-generated scripts. Mentored engineers on production debugging (Valgrind/ASan/TSan/UBSan) and on writing code that survives regulatory audit and stress testing; validated AI-generated CUDA kernels and memory allocation patterns before release.

AI/ML Infrastructure & Agentic Systems Safety Lead at Magical Industries

January 1, 2019 - January 1, 2020

Deployed end-to-end ML and agentic pipelines using SageMaker, Kubeflow, Dataflow, and Docker across AWS/GCP/Azure, serving as final reviewer for AI-generated configurations and deployment scripts. Developed C++/Java real-time market-data ingestion services integrated with Python ML backends, targeting <5ms latency while enforcing strict language-boundary safety contracts. Built voice-controlled conversational AI applications (Python/Java/JavaScript) on AWS Lambda and GCP Cloud Functions and audited generated code for injection flaws and async race conditions.

Production Systems Engineer & Code Safety Validator at Morgan Stanley

January 1, 2016 - January 1, 2016

Built and maintained C++ and Java analytics engines for cross-asset risk systems, acting as final human reviewer for all production release code changes. Developed Python data pipelines for automated order validation and corrected AI-assisted code generation that introduced subtle race conditions in concurrent stream processing. Provided production support for automated trading systems, diagnosing Heisenbugs, memory leaks, and concurrency failures across mixed C++/Python/Java codebases under peak load, including issues that passed basic tests and looked correct to junior reviewers.

Systems Engineering Director & AI Code Validator at Guggenheim Partners / Mezocliq

January 1, 2016 - January 1, 2019

Engineered Python deep-learning models for structured-product analytics and personally reviewed C++ inference runtimes and Python bindings for memory safety prior to production deployment. Designed reinforcement-learning frameworks with custom guardrails preventing unbounded recursion and other risky behaviors from AI-generated strategies. Architected NLP pipelines for real-time sentiment analysis and validated generated JavaScript/Python service code for injection vulnerabilities and async race conditions.

Quantitative Systems Engineer & Data Safety Lead at Federal Reserve Bank of New York

January 1, 2015 - January 1, 2016

Designed secure analytics and data frameworks (Python/C++) for regulatory market-data processing with strict input validation and memory-safety requirements. Built streaming CEP analytics in Java and reviewed template-based/generated Java code for thread-safety and resource leaks. Developed ML/regression methods for predicting missing data and correcting anomalies, validating numerical stability and safe memory usage for all model implementations.

Senior Systems Engineer & Production Debugger at JP Morgan Chase, Bank of America, S&P, Barclays Capital, Greenwich Capital Markets, GAT/BARRA

January 1, 1995 - January 1, 2015

20+ years of C/C++ systems development for low-latency pricing and high-frequency distributed systems, serving as final code-review authority on production releases. Specialized in catching undefined behavior, implicit conversions, buffer overruns, and race conditions that passed compiler checks and unit tests. Implemented Monte-Carlo and lattice-based numerical engines with zero tolerance for memory leaks and non-deterministic floating-point behavior under concurrency. Led emergency debugging and identified single-line causes of segfaults or deadlocks under load, including issues from AI-assisted or junior-written code. Designed/optimized ML-heuristic and mixed-integer programming systems with strict validation to prevent unsafe memory access and unbounded loops.