Shahbaz Shah

Ai Engineer

Arch Technologies • Tiempo completo

Oct 2025 - Feb 2026 • 4 mos

've been working as a developer at Arch Technology for the past 6 months, where I've been focused on building intelligent voice agent systems with RAG-based chat capabilities. This role has given me hands-on experience with cutting-edge AI technologies and real-world conversational AI applications." Project Overview 1 "My main project involves developing a voice-enabled conversational agent that uses Retrieval-Augmented Generation, or RAG. In simple terms, this system allows users to have natural voice conversations with an AI that can pull accurate, up-to-date information from our company's knowledge base in real-time. The challenge we're solving is giving customers instant, accurate answers through voice interaction rather than making them search through documentation or wait for support staff." Technical Implementation "On the technical side, I work across several components: Voice Processing: I integrate speech-to-text services to convert user voice input into text, and text-to-speech for natural-sounding responses. This involves handling audio streaming, managing latency, and ensuring clear transcription even with background noise or accents. RAG Architecture: The core intelligence uses a RAG pipeline. When a user asks a question, we first convert it into embeddings, then search our vector database to retrieve the most relevant documents or information chunks. These retrieved contexts are then fed to a large language model along with the user's query to generate accurate, grounded responses. Chat System: I've built the conversation management layer that maintains context across multiple turns, handles follow-up questions, and ensures the agent stays on topic. This includes managing conversation history, user session state, and implementing safety guardrails. Integration: The system connects to various data sources - internal databases, documentation repositories, and APIs - to ensure the agent has access to current information."