Akshay Dubey

• Architected multi-vector RAG pipeline using OpenAI text-embedding-3-large (3072 dimensions) with 5 semantic dimensions (skills, role, domain, seniority, full-profile) stored in pgvector for candidate-job matching. • Built multi-agent LLM Council orchestrating GPT-4, Claude, and Gemini via OpenRouter with 3-stage process: parallel responses, peer ranking with anonymous labels, and chairman synthesis for consensus answers. • Implemented tool-enabled AI agents using MCP (Model Context Protocol) with OpenAI function calling, supporting iterative tool execution (up to 5 rounds) for complex reasoning tasks. • Designed weighted semantic search with configurable similarity scoring (skills 35%, role 35%, domain 20%, seniority 10%), reducing candidate screening time by 80% across 300+ enterprise organizations. • Integrated Langfuse for LLM observability including prompt tracing, token usage tracking, and latency monitoring across all AI operations. • Built async resume import pipeline with PG Boss queue for multi-stage LLM processing (Parse, Deduplicate, Score) with fuzzy matching and configurable thresholds.

• Built 5+ scalable React applications with TypeScript, delivering responsive interfaces optimized for performance across devices. • Optimized application performance by 20% using React memoization, lazy loading, and code splitting techniques, improving Core Web Vitals scores. • Implemented Redux and RTK Query for efficient state management and API caching, ensuring consistent data flow across components. • Developed reusable component library integrated with RESTful APIs and JWT authentication for secure user access. • Translated designs and wireframes into responsive, high-quality code while collaborating with cross-functional teams. • Conducted code reviews and analyzed user feedback to optimize UI/UX, improving customer satisfaction by 15%.