Challenge
Make a single chat agent an expert across many products and marketing disciplines without bloating every request with thousands of tokens of static prompt context.
AI · Full Stack
Lumen — AI Marketing Agent
Lumen is a modular, context-aware marketing engine built with Next.js. Instead of a single static prompt, a Dynamic Context Engine reassembles the AI's system prompt on every message: a heuristic detector scans the conversation history backwards, identifies which product and marketing skill the discussion is about, and injects only the matching product knowledge and skill playbooks — keeping the agent expert-grade without wasting thousands of tokens. Users switch between LLM providers (DeepSeek, Gemini, Groq Llama 3.1, GPT-4o mini) from the UI, responses stream in as Markdown, and the agent can emit structured payloads that render as interactive multiple-choice forms whose answers feed straight back into the conversation. Threads persist locally with cross-tab synchronization.
Overview1 / 2
Solution
Engineered a heuristic context detector that scans chat history backwards to load only the relevant product knowledge and skill instructions into the system prompt, paired with a model router across four LLM providers and a generative UI layer that renders interactive questionnaires from streamed structured payloads.
Outcome
A token-efficient marketing engine that behaves like a domain expert in every conversation, produces framework-driven strategies (ICE scoring, AARRR funnels), and replaces long typed answers with one-click interactive forms.