Work

Problem

Research becomes difficult to repeat when an experiment depends on loosely connected tools and unstated assumptions. The platform brought research and backtesting into a structured workflow that people and AI agents could operate.

My contribution

I led the design and implementation from concept to production: the research workflows, typed interfaces for agent tools, and validation checks around experiments. The work combined Python engineering with the design of agent-facing interfaces, including MCP access.

Engineering decisions

Explicit inputs and typed interfaces gave each workflow step a clear contract. Repeatable experiment workflows and validation made results easier to inspect and review. The aim was to connect a research idea to an experiment that could be run again and understood.

Scope

Concept-to-production work on agentic research and backtesting infrastructure.

Agentic systems skills in practice

• Tool interface design: typed interfaces made research capabilities usable by both people and AI agents.
• MCP integration: exposed research capabilities through structured agent tools and MCP access.
• Workflow engineering: built repeatable research and backtesting workflows in Python.
• Experiment validation: added checks so experiment inputs and results could be inspected and reviewed.

Problem

Financial learning problems can have limited labeled examples. My research focused on deep-learning methods for this setting, alongside quantitative research using fundamental, price-volume, and alternative data.

My contribution

Developed proprietary deep-learning methods as a Quantitative Researcher. This work sits at the intersection of machine-learning research and financial data.

Problem

Sales workflows involve finding relevant leads, understanding their context, and preparing a useful pitch. Applying LLMs required connecting those tasks into a workflow rather than stopping at a generated answer.

My contribution

I led the Sales Copilot work on lead discovery and pitch generation, building the LLM-based workflows. In a related workstream, I developed machine-learning systems for partner recommendations.

Engineering emphasis

Applied generative AI, recommendation systems, and turning a business task into a concrete machine-learning workflow.

Problem

Large internal search systems need to interpret queries, rank useful results, and support experimentation on changes to search behavior.

My contribution

Worked on ranking, query parsing, and an experimentation framework for internal search as part of Surveillance Analytics.