a
aqil_awan

Aqil Awan

@aqil_awan

LLM Engineer Agentic AI Multi Agent Systems

Pakistán
Inglés
Parte de la información aparece en idioma inglés.
Sobre mí
LLM Engineer with 4+ years of experience building RLHF-trained agentic AI systems, multi-agent RAG architectures, and production-grade LLM applications for government and enterprise clients. Specializes in reinforcement learning (PPO), tool-use policy optimization, semantic retrieval at scale (PostgreSQL/PGVector, FAISS), and multi-agent orchestration (LangGraph, CrewAI, MCP). Track record of shipping systems evaluated with rigorous, quantified metrics (RAGAS, task success rate, tool precision).... Lee más

Habilidades

a
aqil_awan
Aqil Awan
desconectado • 

Revisa mis servicios

Implementación y despliegue de IA
I will build custom ai agents, rag systems, and llm applications
DeUSD 50 / proyectoMás información

Experiencia laboral

Ministry_of Defense

LLM Engineer

Ministry of Defense • Tiempo completo

Feb 2026 - May 2026 • 3 mos

RLHF-Based Agentic AI for Military Decision Support • Developed an RLHF training pipeline using Proximal Policy Optimization (PPO) to improve tool selection policies for AI-assisted military decision support systems while reducing incorrect or unnecessary tool invocations by 31 %. • Designed custom reward functions to optimize tool invocation accuracy, agent decision quality, and multi step operational reasoning. • Evaluated agent performance using task success rate, tool precision, and failure analysis to iteratively improve reliability, safety, and decision consistency. • Technologies: Python, PyTorch, Hugging Face Transformers, TRL, PPO, RLHF, MCP. Multi-Agent RAG Knowledge Assistant • Built a multi-agent RAG assistant indexing 40 vehicle manuals (10,000+ pages, 35,000+ chunks) with PostgreSQL/PGVector, achieving 0.89 Answer Relevancy and 0.91 Faithfulness (RAGAS). • Implemented document ingestion, semantic retrieval, reranking, Arize Phoenix evaluation, and FastAPI deployment. (Crew AI, Ollama, PostgreSQL, PGVector, FastAPI). Multi-Agent Legal Decision Support System • Built an AI-powered legal reasoning system of Military Law that simulated plaintiff, defence, and judge agents to generate evidence-based legal arguments and reasoned judicial decisions using Retrieval-Augmented Generation (RAG). • LangGraph-based multi-agent workflow with FAISS-powered semantic retrieval. Integrated Open AI API for low-latency inference and orchestrated multi-step reasoning across specialized AI agents. (LangGraph, FAISS, Groq API, RAG, TypedDict)