SEMICONDUCTORS — AI News Platform

Overview

SEMICONDUCTORS is an AI-powered analytics platform for semiconductor industry news. It aggregates news from multiple sources and gives the client sentiment and volume charts, a RAG-based chatbot for research queries, daily summaries, and a searchable news feed with filters. Every article has its own page where the client can add tags and labels that feed back into model training.

It was built for a single institutional client. The project started from the foundations of PUMP and then pivoted entirely toward news and research workflows.

My role

Solo full-stack developer. I took this from zero to MVP and owned both the backend and the frontend.

• Designed and implemented the whole system architecture
• Built every frontend feature in Nuxt 3
• Developed the FastAPI backend services
• Created and maintained the Prefect ETL pipelines for news parsing and labeling
• Implemented token streaming for the chatbot with Langchain
• Worked with an ML engineer on the RAG system: he handled the LLM prompts, I helped integrate the Weaviate vector database with async engine support

Key features

News feed and search

• Semiconductor news aggregated from multiple sources
• A feed you can filter and search
• Individual news pages with a tagging UI that feeds model training

Analytics charts

• News volume and sentiment trends over time
• Time-series analytics tied to semiconductor themes and tickers

RAG chatbot

• Research assistant built on retrieval-augmented generation
• Weaviate as the vector database for semantic search
• Real-time token streaming through Langchain
• Answers that blend recent news with company knowledge

Daily summaries

• LLM-generated market digests
• Structured briefings covering themes, company events and industry developments
• Written to work as a morning briefing for the research team

Technical implementation

Backend

• FastAPI for the REST APIs (news, insights, chat endpoints)
• Weaviate vector database for RAG semantic search
• Langchain for LLM orchestration and token streaming

Frontend

• Nuxt 3 in SPA mode, with the Vue 3 Composition API
• Vuetify component library
• Responsive layouts for charts, feeds and the chat interface

Data pipeline

• Prefect for ETL orchestration
• Custom parsers for several news sources
• Automated flows for ingestion, cleaning, LLM summarization and tagging
• Scheduled jobs so the daily summary actually arrives every day

Timeline

Tech stack

Python

FastAPI

Prefect

ClickHouse

PostgreSQL

Nuxt

Vue

OpenAI

Azure

Redis

• Backend: Python, FastAPI, Prefect, Weaviate, Langchain
• Frontend: Nuxt 3, Vue 3, Vuetify
• Infrastructure: ClickHouse, PostgreSQL, Redis, Azure