I'm an AI Engineer working on agent architecture and LLM systems, with a product-focused and practical approach.
I build the systems around language models that help intelligent products move from an idea to something useful.
Off the clockCars. Anything with an engine gets my attention, which might be why I like systems that do more with less fuel.
- Agent architecture
- LLM systems
- Tool calling
- ReAct loops
- LangGraph
- Product-focused
LLM-powered conversation infrastructure for customer-facing AI agents.
- The trace of each step appears here.
Tokens this conversation
A scripted walk-through of how a Flow 2.0 turn is structured. The clients, tools and data are fictional, intents are matched with keywords, and no language model is called. Token counts are illustrative, scaled to the measured 60% reduction.
- Problem
- An AI conversation engine for customer-engagement platforms, where the previous implementation burned far more tokens per conversation than the job needed.
- Approach
- One LangGraph graph (Router, Brain, Tools, Output). The Brain runs a native tool-calling ReAct loop, and the architecture stays client-agnostic.
- Result
- 60% lower token consumption compared to the old implementation.
- My part
- I worked on the agent architecture and the systems that connect language models to useful product actions.
From intentto action
I work on the systems that connect language models to useful product actions while keeping the experience practical and product-focused.
Understand
Conversation contextThe system identifies what a person is trying to accomplish and keeps the interaction grounded in the product.
output / intentDecide
Agent architectureThe model chooses the next useful action within the capabilities and boundaries of the product.
output / next actionAct
Product systemsThe system carries out the action and brings the result back into context for the next decision.
output / product resultRespond
Customer experienceThe completed work becomes a clear, useful response for the person on the other side.
output / useful reply
3rd place
Finovate hackathon
Finance news with an AI agent on board: it helps people plan their money, then lets them run the numbers.
- Read
- A finance news site that keeps people up to date.
- Plan
- A built-in AI agent that helps users plan their finances.
- Calculate
- Calculators for loans, savings and more.
The plan appears here: a debt-free date, a payoff chart you can scrub, and a monthly budget split.
A scripted recreation of Finovate's planning agent: rule-based parsing and a deterministic simulation, with no language model involved. Nothing you type leaves your browser. Illustrative only. Not financial advice.