AI Development
AI development proceeds with data as its essence. The quality and quantity of this data significantly influence the accuracy of AI. However, the accuracy of AI tends to degrade over time. To address this, continuous monitoring of the model and data, as well as periodic retraining, are indispensable. We deeply understand this and fully incorporate the concept of MLOps. While recognizing that data is perishable, we automate the monitoring and retraining of models to consistently offer peak performance.
Difficulties in AI development
AI development works around raw data as its core. Unlike regular system development, the system behavior changes depending on the data, which increases complexity. In AI systems, the proportion of machine learning models (ML Code) is very small, and peripheral elements are enormous and complex.
What Traffine can do
We support AI development considering operational phases, not only data science fields like machine learning modeling. With the awareness that data is raw, we build systems to automate model accuracy monitoring, retraining, and provide the best performance continuously.
Model accuracy monitoring
Data quality monitoring
Example Project ❶
Project
歩行者追跡アルゴリズムの開発
Description
Multiple Object Tracking(MOT)という分野の技術を使い歩行者の検出および追跡を行いました。その際に、他の物体と重なり途切れてしまった歩行者IDを抽出して補正するアルゴリズムを開発しました。
Stacks
Example Project ❷
Project
LLM(大規模言語モデル)およびベクトルデータベースを活用したチャットシステム開発
Description
OpenAI APIとベクトルデータベースのPineconeを組み合わせることで、最新ドキュメントを瞬時に参照できるチャットボットを開発しました。ユーザーからの質問や入力は、OpenAI APIを使用してリアルタイムでベクトル化され、変換されたベクトルはPineconeのデータベースを参照し、類似性の高い回答やドキュメントを瞬時に取得します。ドキュメントの更新情報や新しい知識をチャットボットがすぐに取得して、ユーザーの質問に即座に答えることができます。
Stacks
Example Project ❸
Project
Development of an application to detect the difference between objects in two images.
Description
Utilizing technologies such as object detection, background removal, and feature vector extraction, we developed an application that detects the difference between objects in two images.