The requirement was to build a reference content generation platform that could automatically create high-quality, platform-ready posts (e.g., LinkedIn, Twitter) in the user’s own voice. The client needed the system to source topics from external links such as blogs, articles, and YouTube videos, while mimicking the user’s style using uploaded documents or past content links. They also wanted the ability to pull from the user’s own knowledge base to reflect their expertise, enable refinement loops through interactive prompts, and provide one-click publishing to multiple platforms. Beyond generation, session management, RAG-based document retrieval, and multi-platform support were mandatory for end-to-end usability
CLIENT’S REQUIREMENT
The requirement was to build a reference content generation platform that could automatically create high-quality, platform-ready posts (e.g., LinkedIn, Twitter) in the user’s own voice. The client needed the system to source topics from external links such as blogs, articles, and YouTube videos, while mimicking the user’s style using uploaded documents or past content links. They also wanted the ability to pull from the user’s own knowledge base to reflect their expertise, enable refinement loops through interactive prompts, and provide one-click publishing to multiple platforms. Beyond generation, session management, RAG-based document retrieval, and multi-platform support were mandatory for end-to-end usability
Solution
We built a platform where multiple agents powered by LangGraph and LangChain work together to generate content that is both personalized and platform-ready. The process begins with a Query Parsing Agent that reads the user’s request and decides what needs to be done, whether to extract content, apply style matching, or bring in user references.
If style is required, the request goes to a Style Optimization Agent, which uses uploaded documents or past posts to capture the tone and writing patterns. At the same time, the system triggers an Extract Content Agent that pulls transcripts from YouTube links or text from blogs and articles. For cases where users want their own expertise reflected, a Document Retrieval Agent uses RAG (retrieval augmented generation) to fetch relevant notes, case studies, or uploaded files from the user’s knowledge base. These parallel outputs, style, topic content, and personal references, are then passed to the Draft Generation Agent, which uses large language models to merge them into a first draft. The draft is shown to the user, who can refine it by adding prompts, new links, or stylistic tweaks, and the system dynamically re-runs the agents to update the content.
Once the user approves, a Publishing Agent posts the final content directly on platforms like LinkedIn, returning the live URL. In this way, the system handles the full cycle, query understanding, multi-source content extraction, RAG-based personalization, iterative refinement, and seamless publishing, making the workflow both intelligent and automated.
Technology Stack
Frontend : Steamlit , LLM: deepseek-r1:8b switchable between Gemini 2.0 Flash (local & cloud LLM), backend : Flask, Docker, DB : Chroma DB (vector), Agentic FW: Langchain, Lang graph, Lang smith, Lang chain Olama embeddings: nomic-embed-text
