Download: Styled PDF · Plain-text PDF · Plain text (.txt)
Independent Consultant in Agentic AI, Scientific Computing, and Quantitative Systems Architecture
Contact: augusto [dot] kiel [at] gmail [dot] com
Executive Profile
Scientific Architect with a Physics background and elite tenure across Tier-1 financial institutions and scientific R&D organizations (Qontigo, SimCorp, Mercado Libre, AstraZeneca). I specialize in Research Software Engineering (RSE), High-Performance Computing, and Production-Grade AI Engineering. My work focuses on translating complex mathematical models into stable production systems, ranging from Neural Stochastic Differential Equations (NeuralSDEs) and Domain-Specific Languages (DSL) to Agentic Orchestration and Evaluation Frameworks (Evals). I help firms move research-quality models into production without losing mathematical correctness or computational performance.
My Physics degree (Licenciatura, University of Buenos Aires) taught me to find the right level of abstraction before writing a line of code. Seven years across JP Morgan, Qontigo, SimCorp, and Mercado Libre taught me what production actually costs.
Core expertise: High-Performance Computing • Neural Stochastic Differential Equations • Agentic AI platforms • Multi-asset pricing systems • Risk analytics • DSL development • Numerical optimization • Python/C#/Julia • System Architecture
Engagement Focus: Performance Optimization Sprints, Agentic Platform Integration, Knowledge Engine Architecture, Agent Evaluation Frameworks, RSE Consulting, DSL Design, Architectural Design Sprints, Specialized Corporate Training.
Consulting Services
Agentic AI Engineering
- Production-grade Model Context Protocol (MCP) server design and implementation
- Multi-tenant enterprise authentication architecture (OAuth 2.0 PKCE, Grant Chains, CCG)
- Competitive Intelligence Knowledge Engines: document ingestion pipelines, hybrid RAG, IR trace optimization
- Automated Agent Evaluation (Evals) frameworks for task accuracy, tool-calling reliability, and retrieval stability
- Multi-layer agent memory systems (episodic, semantic, working memory) and personalization engines
- LLM security architecture and threat modeling for advanced reasoning models
- Agentic cost engineering: pre-filtering, caching, and execution graph optimization
Research Software Engineering (RSE)
- Translating research-quality code into stable, reproducible production systems
- Neural Stochastic Differential Equations (NeuralSDEs) and physics-informed ML for scientific simulation and quantitative finance
- Standardizing analytical workflows from Jupyter notebooks to CI/CD-managed pipelines
- Reproducibility, testing, and DevOps practices for R&D and Quantitative Finance teams
- Scientific computing architecture for high-mathematical-complexity domains
- Migration and modernization of legacy analytical systems
High-Performance Computing (HPC) Sprints
- Performance audit and optimization for numerical bottlenecks
- Vectorization, caching, and parallelism strategies
- 300% performance gains achieved in production systems
- Real-time computational finance optimization
- Multi-asset class pricing and valuation frameworks
Domain-Specific Language (DSL) Architecture
- Custom language design for complex pricing models
- Interactive UI development for quantitative libraries (ipywidgets, Voilà)
- Simplifying access to sophisticated mathematical models for non-engineer users
Training & Knowledge Transfer
- High-Performance Numerical Computing with Julia workshops
- Research Software Engineering (RSE) best practices for scientific teams
- Corporate workshops on quantitative finance and computational methods
- Technical mentorship for quant and engineering teams
- Interactive educational materials using Jupyter notebooks
Education
Licentiate degree in Physics, University of Buenos Aires (2011-2017)
Signature Case Studies
1. The Performance Optimization Sprint (Qontigo)
The Challenge: Critical risk calculations were too slow for real-time reporting due to unoptimized convertible bond pricing engines.
The Solution: Led a forensic performance audit and implemented cache optimization strategies in the core C# numerical library.
Outcome: Achieved a 300% performance gain, enabling real-time production risk reporting and significantly reducing Azure compute spend.
2. The Scientific AI Implementation (Research/SimCorp)
The Challenge: Traditional Monte Carlo simulations for European Option Pricing were computationally expensive; standard AI lacked mathematical constraints.
The Solution: Managed research into Neural Stochastic Differential Equations (NeuralSDEs) using Julia, combining deep learning with physical laws. Parallelly developed an LLM-based RAG system for querying complex financial documentation.
Outcome: Demonstrated superior convergence speeds over traditional solvers and established a framework for “Safe AI” in financial contexts.
3. The “Quant Experience” Architecture (SimCorp)
The Challenge: Quants struggled to interact with complex underlying pricing models, leading to errors and slow iteration.
The Solution: Designed and developed a Proof-of-Concept Domain-Specific Language (DSL) and integrated interactive Jupyter-based UIs (Voila/ipywidgets).
Outcome: Enabled non-engineers to safely construct and test pricing logic by abstracting underlying complexity.
4. The Enterprise Agentic Platform (AstraZeneca, Oncology R&D)
The Challenge: An R&D agentic platform needed a complete external-integration surface, a measurable quality framework, and reliable knowledge retrieval — none of which existed.
The Solution: Architected production-grade MCP servers with multi-tenant OAuth patterns; built automated Evals frameworks to quantifiably measure agent task accuracy and tool-calling reliability; designed document ingestion pipelines and optimized IR traces for a Competitive Intelligence Knowledge Engine; engineered multi-layer agent memory and personalization systems.
Outcome: Delivered the full external integration layer to production; established reusable OAuth blueprints adopted for all subsequent integrations; reduced LLM token consumption by 13%+ through semantic pre-filtering and cache optimization; established an Evals suite tracking tool-calling accuracy and retrieval precision as ongoing KPIs.
Professional Experience
AstraZeneca
Senior AI Engineer (Independent Contractor) March 2026 - Present | Remote
Advancing the R&D Agentic Platform and strengthening internal developer enablement for agent-based workflows.
- Enterprise Agentic Integrations: Architected and shipped the external integration surface of the R&D agentic platform, designing production-grade Model Context Protocol (MCP) servers with robust multi-tenant authentication patterns (OAuth 2.0 PKCE, Grant Chains, and Client Credentials).
- Competitive Intelligence & Knowledge Engines: Built automated document ingestion pipelines to index enterprise sources into the LLM-Wiki/Knowledge Base. Tuned hybrid search (dense + sparse) and analyzed IR traces to cut retrieval latency and improve source hit rate.
- Automated Agent Evaluations (Evals): Designed and ran automated evaluation frameworks measuring agent task-execution accuracy, tool-calling reliability, and retrieval stability across model updates.
- Multi-Layer Memory & Personalization: Built user content personalization engines and structured agent memory systems—episodic, semantic, and working memory layers—so agents return context-relevant responses rather than generic ones.
- LLM Security & Threat Modeling: Audited tool-use capabilities in advanced reasoning architectures, identifying and neutralizing critical tool-injection vectors and token-abuse vulnerabilities before execution in production.
- Agentic Cost Engineering: Optimized execution graphs, achieving a 13%+ reduction in LLM token consumption through semantic pre-filtering and custom cache optimization layers.
Phorma
Co-founder & Research Software Engineer February 2026 - Present | Buenos Aires, Argentina
Co-founded Phorma with Agustín Corbat to apply Research Software Engineering (RSE) to R&D and Quantitative Finance teams. Scientists keep their research focus; Phorma owns the engineering execution.
- Transforming research-quality models and workflows into stable, reproducible production systems
- Designing RSE architectures for scientific simulation and quantitative finance
- Establishing engineering best practices (testing, CI/CD, reproducibility) in high-mathematical-complexity codebases
Independent Consultant
Quantitative Software & Scientific Computing January 2026 - Present | Buenos Aires, Argentina
Providing specialized consulting services to financial institutions and technology companies:
- Quantitative finance system architecture and development
- Scientific computing and numerical software optimization
- Technical leadership and team mentorship
- Developer experience and tooling for analytical platforms
Focus areas: Multi-asset pricing libraries, risk analytics systems, DSL development, performance optimization, Python/C#/Julia consulting.
Mercado Libre
Software Technical Lead, IT Staff / Financial Planning & Analytics June 2025 - March 2026
Leading technology strategy and managing 14 engineers across Financial Planning & Analytics for Latin America’s largest e-commerce ecosystem.
Key Focus: Strategy, Standardization & AI Workflows
- Architecting scalable financial planning and analytics platforms
- Championing AI-assisted development workflows and clean architecture standards
- Transforming ad-hoc Jupyter analyses into CI/CD-managed production systems
Impact: 90% reduction in forecasting pipeline errors through RSE principles; 15% velocity increase across engineering team
Technologies: Go, TypeScript, Python, BigQuery, Jupyter, CI/CD, distributed systems
SimCorp
Lead Software Engineer, Core Analytics March 2024 - May 2025 | 1 year 3 months
Key Focus: Core Analytics & Quant UI
- Developed Domain-Specific Language (DSL) POC for pricing model interaction
- Created LLM-based RAG system for financial documentation Q&A
- Integrated Quant UI with Axioma Risk UI for institutional investors
- Redesigned libraries for Automatic Differentiation (AD) support
Qontigo (Axioma Risk)
Associate Principal, Core Analytics September 2020 - March 2024 | 3 years 7 months
Key Focus: Core Quant Libraries & High-Performance Computing
- Designed and developed core Quant Monorepo (C#) for Analytical Libraries
- Managed NeuralSDE research internship for European Option Pricing using Julia
- Built Axioma Pricing Library (APL) from ground up with 100% accuracy
- Developed comprehensive curve construction library (rates, yields, discounts, spreads)
- Led development of interactive UI (ipywidgets/voila) for Quant library access
Impact: 300% performance gain on Convertible Bond Pricing Engine enabling real-time production calculations; “Exceptional Performance” rating (2023)
J.P. Morgan
Technology Analyst, Rates CIB July 2018 - August 2020 | 2 years 2 months
- Built production-grade financial reporting systems with zero-downtime requirements for Rates CIB
- Provided critical support to Rates Quant team, enhancing analytical capabilities
- Delivered infrastructure for mission-critical reporting services ensuring compliance and reliability
Open Source Contributor & Content Creator
Scientific Software & Educational Tools February 2016 - Present | 9+ years
@akielbowicz | YouTube: @SCA314 | GitHub: SCA314
- Creator of SCA314, an educational YouTube channel focused on software craftsmanship, scientific computing with Julia, and automated testing practices in Spanish
- Development of interactive educational materials based on Jupyter notebooks
- Contributions to scientific computing and data visualization projects
- Created charly-vibes (site, GitHub) as a personal R&D initiative to explore the capabilities and limits of AI-assisted coding and agentic autonomous development across platforms
- Educational content bridging academic knowledge and industry best practices
Teaching Experience
University of Buenos Aires
- Professor of Calculus and Linear Algebra, CBC Engineering (December 2020 - July 2022)
- Teaching Assistant on Summer Course of Optics and Thermodynamics for Biology and Geology (February 2015 - March 2015)
- Science Communicator at Physics Department (March 2013 - December 2014)
Southern International School
- High School teacher of Physics, Mathematics and Information Technologies (2016)
Publications
Shared Memory Semi-Implicit Solver for Hydrodynamical Instability Processes (2023)
Photon Counting Module based on Avalanche Photo-Diodes (2017)
Speaking & Community
Active participant in technology conferences and meetups as speaker and collaborator:
- SciPy Latinoamérica 2022 (Argentina): Workshop presenter
- Regular speaker at Python and Julia user groups in Buenos Aires
- Represented Qontigo at ECI UBA (School of Information Sciences, University of Buenos Aires)
All talks available at: talks.saxa.xyz
Technical Skills
Programming Languages Production: Python, C#, Julia Functional/Niche Languages: F#, Clojure
Agentic AI: Model Context Protocol (MCP), LangChain, LangGraph, LlamaIndex, RAG pipelines, vector databases, AWS Bedrock, multi-agent systems, LLM Evals, enterprise OAuth (PKCE, OBO, CCG)
Research Software Engineering: Reproducible workflows, scientific Python stack, numerical methods, automatic differentiation, stochastic differential equations, NeuralSDEs
Quantitative Finance: Multi-asset pricing, risk analytics, derivatives valuation, curve construction, model calibration
Architecture & Leadership: DSL design, API design, monorepo infrastructure, microservices, Technical Mentorship, Solution Design, Corporate Training
Tools: Jupyter, Git, Docker, Azure, GitHub Actions, Visual Studio
Available for consulting engagements globally (remote) and in Buenos Aires
Contact: augusto [dot] kiel [at] gmail [dot] com