Loading...
Loading...
No more manual customs code lookup. AI finds the right code in seconds.
Tariff8AI is a business-focused AI product designed to reduce the time and uncertainty involved in customs classification. The challenge was not only to deliver accurate predictions, but also to make the output explainable and usable in a real operational workflow.
I shaped the product logic, translated classification rules into a usable decision flow, and built the AI-enabled application stack end-to-end — from backend APIs and retrieval architecture to the user-facing workflow.
Classification reduced from hours to seconds
92%
From hours to seconds
10,000+ rules
Customs classification is a high-friction process: the information is often incomplete, the rules are dense, and decisions need to be documented. A generic chatbot would not be enough because the output needed to be reliable, structured, and explainable.
The product had to support a real business workflow where users search, compare possible classifications, and understand why a given code is recommended.
I designed an AI-assisted classification workflow combining FastAPI, LlamaIndex, Qdrant, and OpenAI models. The system indexes customs knowledge, retrieves the most relevant passages, and produces a recommendation with supporting rationale.
On top of the retrieval and classification layer, I built a product flow that helps users move from question to decision quickly, instead of exposing raw model output.
The main engineering challenge was balancing accuracy, latency, and traceability. The solution had to combine semantic retrieval with explainable backend responses, rather than opaque one-shot generation.
This led to an architecture where retrieval, prompt construction, and response formatting were designed as explicit steps, making the system easier to improve, validate, and reason about over time.
Let's talk about your product, platform, or operational workflow and see how I can help shape and build it.