AGI Ready platform

OVERVIEW We are launching a highly ambitious, bootstrapped AI intelligence and discovery platform. This is not a static directory or a simple database. The goal is to build an active, evolving ecosystem platform that dynamically tracks, maps, and predicts the trajectory of the AI industry by analyzing three core pillars: -Top AI Researchers, -Top AI Patents, and -Top AI Startups. The platform will cross-reference these pillars to uncover hidden insights (e.g., matching a new patent to a startup's hidden trajectory, or tracking a researcher's spin-off venture). Because this is a self-funded, bootstrapped project, extreme efficiency is our highest priority. We need an expert engineer who can build highly intelligent, self-optimizing multi-agent workflows *without* causing massive LLM token bills. THE VISION: BEYOND A DIRECTORY We want this platform to act as a living intelligence engine. It needs to include: -Dynamic Knowledge Graphing: Automatically connecting researchers to patents, patents to startups, and startups to funding/compute resources. -Autonomous Discovery & Synthesis: Agents that don't just scrap data, but actively synthesize *why* a patent or research paper matters, generating automated deep-dives and market intelligence reports. -Recursive Self-Improvement (RSI): An architecture built for "AGI-readiness." The data pipeline must run on an iterative reasoning loop where agents review their own data gaps, autonomously write new search/retrieval queries, and continuously refine their own prompts and validation logic to improve data hygiene over time. KEY RESPONSIBILITIES -System Architecture: Build a highly scalable, modular backend (Python/FastAPI preferred) capable of handling asynchronous multi-agent orchestration. -Frontend UI/UX: Create a clean, fast data-visualization frontend (Next.js/Tailwind) that showcases trends, connections, and automated insights seamlessly. -Ultra-Lean Agentic Pipeline: Implement aggressive token-saving strategies (prompt caching, semantic caching, structured outputs, and routing heavy tasks to low-cost models like GPT-4o-mini / Claude 3.5 Haiku). -Self-Correcting Loops: Design the background RSI/agentic workflow that continuously audits database quality, identifies missing data links, and self-corrects without human intervention. TECH STACK PREFERENCES -Backend: Python (FastAPI) -Agent Framework: LangChain, LlamaIndex, CrewAI, or a custom lightweight agent loop (to minimize overhead) -Database/Graph: PostgreSQL/Supabase (with pgvector) or a lean Graph Database solution Frontend: Next.js / Tailwind CSS REQUIREMENTS -Proven experience building complex, multi-agent AI systems and autonomous reasoning loops. -Deep knowledge of LLM cost-reduction techniques and context-window management. -Experience with graph structures or complex relational data mapping. -A "hacker" mindset—finding brilliant, low-compute architectural workarounds to minimize API token consumption. HOW TO APPLY Please start your proposal with the phrase "Intelligence Engine" so I know you read the whole post. In your application, briefly answer: 1. How would you architect a self-improving (RSI) data loop so that it doesn't get stuck in an expensive, infinite token-consuming loop? 2. What is your go-to strategy for minimizing context windows and caching prompts for repetitive agent tasks?