Umairmalik _

Associate AI Engineer with hands-on experience building production-grade Generative AI applications using Python, FastAPI, OpenAI APIs, LangChain, LangGraph, Retrieval-Augmented Generation (RAG), LangSmith, and Hugging Face Transformers. Experienced in developing scalable multi-agent systems, enterprise conversational AI, semantic search platforms, and NL-to-SQL intelligence solutions. Skilled in integrating LLMs with FastAPI services, vector databases, and enterprise APIs while optimizing inference cost, response quality, and system reliability.

Passionate about building intelligent AI solutions that automate complex business workflows.

ASSOCIATE AI ENGINEER - OAK STREET TECHNOLOGIES

(2025-08)

Tech Stack: Python, FastAPI, OpenAI API (GPT-4.1 Mini & GPT-4.x), LangChain, LangGraph, LangSmith, Hugging Face Transformers (DistilBERT), Qdrant, ChromaDB, Weaviate, Azure Key Vault, SQL

• Engineered a production-grade multi-agent AI platform using FastAPI, LangChain, LangGraph, Retrieval-Augmented Generation (RAG), and OpenAI models to support enterprise-scale conversational AI solutions for 1,000+ expected users.
• Designed and developed 10+ FastAPI backend services and AI endpoints powering enterprise conversational workflows, intelligent routing, document retrieval, and NL-to-SQL capabilities.
• Architected a supervisor-agent workflow using LangGraph State Graphs with specialized agents responsible for query understanding, retrieval, SQL generation, report interpretation, validation, and response generation.
• Implemented tool-calling workflows, conversational memory, checkpointing, prompt templates, and human-in-the-loop interactions to improve workflow reliability and maintain conversational context.
• Integrated more than 10 enterprise APIs to enable secure, real-time retrieval of business information across multiple internal systems.
• Designed and implemented an enterprise NL-to-SQL solution capable of translating natural language business questions into schema-aware SQL queries across approximately 70 relational database tables.
• Developed a retrieval-enhanced SQL generation and validation pipeline using OpenAI models, semantic retrieval, metadata-aware context selection, and agent orchestration.
• Achieved approximately 97% SQL generation accuracy through schema-aware prompting, contextual retrieval, and automated query validation.
• Automated reporting workflows across more than 90 enterprise reports, reducing manual reporting effort by an estimated 88–93%.
• Enabled conversational access to reports containing over 80,000 records, reducing information retrieval time from minutes to seconds.
• Optimized LLM inference costs by strategically utilizing GPT-4.1 Mini for cost-sensitive workloads while dynamically leveraging larger OpenAI models for complex reasoning tasks, balancing operational cost with response quality.
• Applied prompt engineering techniques to improve response consistency, retrieval quality, and structured output generation across enterprise AI workflows.
• Utilized LangSmith for observability, request tracing, debugging, and monitoring of LLM execution paths to improve system reliability and simplify troubleshooting.
• Fine-tuned a Hugging Face DistilBERT transformer model for enterprise intent classification using approximately 7,000 proprietary natural language queries spanning multiple business intent categories.
• Improved intent recognition accuracy to approximately 80%, enabling intelligent request routing and more accurate agent selection within LangGraph orchestration workflows.
• Developed an AI-powered enterprise search platform using semantic search, hybrid retrieval, embedding-based search, and metadata filtering across Qdrant, ChromaDB, and Weaviate vector databases.
• Evaluated multiple vector database technologies to compare retrieval quality, scalability, indexing performance, and enterprise search effectiveness before selecting solutions for production use.
• Integrated Azure Key Vault to centralize secure management of API keys, database credentials, and environment-specific configuration across staging, UAT, and production environments.
• Applied object-oriented software engineering principles to develop reusable components supporting secure credential management and enterprise AI services.
• Performed feature validation and functional testing for newly implemented capabilities before integration into the production codebase.

WEB DEVELOPER INTERN - 9D TECHNOLOGIES

(2024-06 - 2024-08)

Tech Stack: JavaScript, React.js, Node.js, Express.js, MongoDB, HTML, CSS

• Developed reusable frontend components and interactive user interfaces using JavaScript, HTML, and CSS for internal business applications.
• Collaborated on the development of a MERN-stack Online Quran Learning platform, contributing to both frontend and backend application features.

BS - Computer Science - FAST NUCES (2021 - 2025)