← Back to work
FPT Software
Agentic Report
An agentic pipeline for Q&A over a Snowflake data platform, plus machine-learning alerts when the data looks abnormal.
- Role
- AI Engineer
- Context
- FPT Software
- Period
- Oct 2025 – Feb 2026
Overview
Developed an agentic application enabling users to query and interact with a large-scale data platform (Snowflake) through natural-language Q&A, and provided data insights through a rebuilt library, semantic layer, and context base.
Analyzed and escalated abnormalities or unexpected incidents to users in near real time using machine learning algorithms.
Key achievements
Built an application that gives the team quick insights into running data flows and high-priority incidents as they occur.
Enabled the Data Analysis team to boost productivity and reduce duplicated workload by capturing and reusing the most frequently used patterns.
Responsibilities
- Designed prompts and agentic pipelines with LangChain, LangGraph, and the Anthropic Deep Agents framework, including a tree-based decision pipeline for complex responses, caching, and cost optimization.
- Developed a semantic layer for better understanding of data structure, applying context engineering combined with expert data-analysis knowledge to adapt to business needs.
- Implemented the backend using Python/FastAPI and deployed it on AWS instances.
- Curated a golden QA dataset of 300+ pairs for evaluating model accuracy and detecting model drift.
- Researched machine learning algorithms, focusing on time-series and statistical methods, to automatically detect abnormalities and incidents from the provided data.
Tech stack
Languages & Frameworks
PythonFastAPILangGraphLangChain
Data Platform
Snowflake
LLMs
Anthropic AI services
Platform
AWS