Ansar Qureshi

Ansar Qureshi

AI Engineer specializing in Generative AI, RAG pipelines, and production-grade LLM applications.

Based in: LahoreCurrently: AI Engineer, Magnus Mage Pvt. Ltd.Experience: 3 years

Hire me

Ansar is an experienced AI Engineer proficient in building scalable machine learning systems, semantic search, and computer vision solutions using Python and LangChain. He has a proven track record of deploying local LLM infrastructures and hybrid security analyzers for real-world business applications.

Experience

AI Engineer

Magnus Mage Pvt. Ltd.workMar 2024 – Present

Summary

Developed and implemented AI/ML and Generative AI solutions for real-world applications, contributing across the complete development lifecycle from experimentation to deployment.

What I did

  • Worked on LLM-based applications, RAG pipelines, embeddings, vector databases, prompt engineering, and AI-driven automation.
  • Designed and integrated AI services and backend APIs, focusing on reliable inference, performance, scalability, and production-oriented deployment.
  • Evaluated and optimized different AI models, inference approaches, and deployment strategies based on application requirements and available computing resources.
  • Built a smart help RAG system for the company website to answer visitor inquiries about company services.
AI/MLGenerative AILLMRAGEmbeddingsVector DatabasesPrompt EngineeringAI AutomationBackend APIsRecursive Character Text Splitter

AI Engineer intern

Magnus Mage Pvt. Ltd.internshipDec 2023 – Mar 2024

Summary

Built and evaluated machine learning solutions for real-world use cases, including Credit Card Fraud Detection, Telecom Churn Prediction, and Library Management, using Python, data preprocessing, feature engineering, model training, and performance evaluation to develop reliable predictive systems.

What I did

  • Analyzed confusion matrices for credit card fraud detection and telecom churn models to identify actual fraud values beyond basic accuracy.
PythonData PreprocessingFeature EngineeringModel TrainingMachine LearningConfusion Matrix

Skills

Technical

PythonPython
Prompt Engineering
NumPy
Pandas
LangChain
ChromaDB
Faiss
FastAPIFastAPI
GitGit
Scikit-learn
PyTorchPyTorch
DockerDocker
DjangoDjango
Matplotlib
Postman
MongoDBMongoDB
MySQLMySQL
PostgresqlPostgresql
ReactReact
HTML5HTML5
CSS3CSS3
JavaScriptJavaScript
Ollama
Recursive Character Text Splitter
Confusion Matrix
vLLM

Projects

LLM-Powered AI Applications

AI Developerprofessional
Designed and deployed multiple LLM-powered applications, including Meal Planner, Workout Planner, Idea Generator, and Smart Budget, using LLaMA 3.1 70B/405B Instruct with tailored prompt engineering to deliver personalized, context-aware, and mathematically accurate outputs.
LLaMA 3.1 70BLLaMA 3.1 405BPrompt Engineering

Local LLM Infrastructure

Infrastructure Engineerprofessional
Development of a local LLM inference infrastructure, evaluating coding-focused models for RTX 5070 (12GB) and RTX 5090 (32GB) deployments using Ollama, with a focus on Mixture-of-Experts (MoE) architectures, quantization, VRAM optimization, and multi-user concurrency. Deployed containerized LLM services using Docker, FastAPI, and Redis, and resolved cross-machine inference and Ollama connectivity issues. Evaluated and integrated NVIDIA NIM, Ollama, llama.cpp, and vLLM for local and production-oriented inference workflows.
OllamaRTX 5070RTX 5090MoEQuantizationDockerFastAPIRedisNVIDIA NIMllama.cppvLLM

AI-Powered Solidity Security Analyzer

Lead Developerpersonal
Built a hybrid Solidity smart contract analyzer combining Regex/AST-based static analysis with LLM inference, implementing confidence-gated routing to determine when deterministic rules or LLM-based analysis should be used. Developed a Solidity-focused AI chatbot with FastAPI, enabling developers to interact with and analyze Solidity smart contracts using LLM-powered assistance.
SolidityRegexASTLLMFastAPI

Date Fruit Ripeness Classification

Deep Learning Researcheracademic
Developed a deep learning-based image classification system using PyTorch to classify date fruits into two categories: Fully Ripe and Not Ripe. Preprocessed and augmented fruit images, trained a CNN-based transfer-learning model, and evaluated its performance using classification metrics to enable automated, image-based ripeness assessment. Designed the classification pipeline with potential smart-agriculture and automated harvesting applications, supporting objective and consistent ripeness detection.
PyTorchCNNTransfer LearningImage Processing

Credit Card Fraud Detection

personal
Developed a model during an internship using Kaggle datasets to identify actual fraud cases by analyzing confusion matrix results.
Kaggle datasets

Telecom Churn Model

personal
Created a churn prediction model during an internship, focusing on specific requirement metrics rather than just overall accuracy.
Kaggle datasets

Education

Bachelor of Science (BS) in Computer Science

The Islamia University of BahawalpurComputer ScienceSep 2019 – Sep 2023
Hire AnsarReach out about a role, a contract or a conversation.For recruiters

Ansar's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis