vector

Vector is a research dashboard that automatically collects cutting-edge overseas tech news, uses AI to translate and summarize it into Japanese, and allows users to check the key points, background, and trends needed for investment decisions. Its goal is to reduce the burden of continuously following English articles and make it possible to grasp "what happened this week" in fields such as AI, semiconductors, and space. Key features include automatic collection of tech news, AI-based translation, summarization, and background organization, category-based article lists and filtering, related article recommendations, weekly LLM briefings, and aggregation of notable and rapidly rising words. It also includes an AI agent function that researches across internally accumulated articles and externally retrieved articles to generate answers to questions. The technology stack uses Next.js 16 (App Router/BFF), React 19, TypeScript, Tailwind CSS v4, and shadcn/ui on the frontend, with Better Auth for authentication. The backend uses Python 3.13, FastAPI, Pydantic/SQLModel, and Alembic. Article processing uses an event-driven architecture with EventBridge Scheduler, AWS Lambda, Amazon SQS, and Transactional Outbox. Data is stored in Amazon RDS for PostgreSQL with pgvector (768-dimensional vector search) and ElastiCache Valkey. For AI, it uses Gemini (translation, summarization, research planning, answer generation, and Embedding) and DeepSeek (importance and investment context analysis, search query generation, and evidence verification), and uses Amazon Bedrock AgentCore Gateway for external search. The production environment runs on AWS (ap-northeast-1), with ALB as the only public entry point, and frontend, API, and worker running on ECS Fargate. Infrastructure is managed with Terraform, and the network boundary is separated into public entry, internal API, external HTML retrieval processing, and DB permissions to minimize the impact scope of processes that handle external input. Locally, it can be started with Docker Compose, and Gemini/DeepSeek API keys and various secrets must be configured. Note that this repository does not currently have an open-source license, and prior permission is required for reuse, modification, or redistribution of the code.