Annas M

AI Engineer

Niblon • Full time

Apr 2022 - Oct 2025 • 3 yrs 6 mos

• Designed and deployed 10+ AI and RAG systems with adaptive memory and reflexive reasoning • Built LLM-as-Judge pipelines and Human-in-the-Loop evaluators, improving accuracy by 25% • Enhanced RAG retrieval precision with embedding quantization, adaptive chunking, and hybrid ranking, reducing response latency by 40% • Applied MLOps practices (monitoring, cost optimization, model iteration) and optimized inference latency through hybrid retrieval and caching. • Developed voice-enabled AI agents using Whisper (STT) and ElevenLabs (TTS), processing 10K+ voice interactions monthly with 92% transcription accuracy • Researched Agent-to-Agent and MCP-based reflexive reasoning for scalable automation • Architected and deployed hybrid RAG systems with multi-step retrieval pipelines, combining dense embeddings (FAISS) and sparse retrieval (BM25) to achieve 35% improvement in answer relevance • Built custom evaluation frameworks using RAGAS metrics to monitor and improve RAG performance • Implemented multi-step reasoning chains with LangChain and custom tool-calling mechanisms, handling complex queries requiring 3-5 sequential API calls • Optimized embedding generation and caching strategies, reducing API costs by 45%