Nuaima

Production AI Engineer specializing in end-to-end ML products and RAG systems.

Based in: PakistanCurrently: AI Engineer, EFANIExperience: 1 year

A BS Artificial Intelligence graduate from GIKI with extensive experience building and deploying production-grade AI systems across web and mobile platforms. Proven track record in reducing manual triage by 40% through LLM pipelines and publishing peer-reviewed research in computer vision and generative AI.

Experience

AI Engineer

EFANIRemote / Lahore, PakistanworkFeb 2026 – Present

Summary

Built, deployed, and operated production AI automation across 9,000+ historical support records and 500+ daily tickets using Python, FastAPI, LLM pipelines, Sentence Transformers, FAISS, RAG, REST APIs, and SQL. Developed ticket classification, priority prediction, semantic retrieval, and intent-routing workflows across 10+ categories, reducing manual support triage by approximately 40%. Built 4 production FastAPI inference services with authentication, RBAC, schema validation, structured logging, enterprise API integrations, error handling, and Dockerized deployment. Improved reliability through confidence thresholds, retrieval evaluation, failure analysis, monitoring, fallback logic, and human escalation for low-confidence predictions.

What I did

Managed Dockerized deployment of ML services.
Implemented supervised classification for ticket category and priority prediction across 10+ classes.

Results

Reduced manual support triage by approximately 40%.
Successfully automated workflows for 500+ daily tickets.

PythonFastAPILLM pipelinesSentence TransformersFAISSRAGREST APIsSQLRBACDockerAuthenticationLoggingValidation

AI/ML Engineer Intern

Analita InfinitiLahore, PakistaninternshipOct 2025 – 2026

Summary

Built generative and multimodal AI workflows using Python, PyTorch, Stable Diffusion XL, Hugging Face Diffusers, CUDA, FastAPI, and OpenCV. Optimized inference through FP16 precision, GPU acceleration, model benchmarking, attention optimization, and GPU memory management. Developed image and video generation pipelines and integrated model inference into application-facing APIs for deployment and product testing.

What I did

Optimized Stable Diffusion XL inference using FP16, attention optimizations, memory cleanup, and reduced tensor retention.
Benchmarked different inference configurations to achieve faster and more stable generation.

Results

Lowered GPU memory pressure and reduced out-of-memory failures during 1024×1024 generation.
Enabled larger image-generation workloads to run reliably on available GPUs without crashing or requiring model restarts.

PythonPyTorchStable Diffusion XLHugging Face DiffusersCUDAFastAPIOpenCVFP16 precisionAttention Optimization

AI Summer Research Intern

University of Europe for Applied SciencesRemote, GermanyinternshipJun 2024 – Jul 2024

Summary

Trained and evaluated computer vision models on 87K+ images across 29 classes using TensorFlow, Keras, YOLOv5, and OpenCV, achieving approximately 95% accuracy. Built end-to-end training workflows covering preprocessing, augmentation, train/validation splits, model training, checkpoint evaluation, and error analysis across large-scale visual datasets.

What I did

Fine-tuned pretrained models including YOLOv5 and TensorFlow/Keras architectures on a custom 29-class dataset with over 87,000 images.
Managed dataset preparation, augmentation, train/validation splits, checkpoint selection, hyperparameter tuning, and error analysis.

Results

Achieved approximately 95% accuracy on computer vision models.

TensorFlowKerasYOLOv5OpenCVTransfer LearningFine-tuning

Skills

Technical

LLMs

Python

RAG

FastAPI

Data Pipelines

REST APIs

PyTorch

Hugging Face

Deep Learning

Computer Vision

Sentence Transformers

Embeddings

FAISS

Docker

Git

Stable Diffusion XL

LangChain

Authentication

PostgreSQL

TypeScript

SQL

RBAC

Structured Logging

Monitoring

Linux

pgvector

TensorFlow

React

Scikit-learn

CLIP

Whisper

Google Cloud

JavaScript

Flutter

Keras

LangGraph

GitHub Actions

AWS

Supabase

CI/CD

C++

Error Handling

Netlify

Firebase

Cloudinary

Stripe

Fine-tuning

Random Forest

XGBoost

SHAP

Supervised Classification

Google Play Billing

Projects

AI Voice & Audio Transcription Platform

Lead Developerpersonal

Built a production-oriented audio pipeline from ingestion to Whisper transcription, timestamped segmentation, and structured API responses using Python and FastAPI. Added health checks, request validation, automated tests, structured output, and deployable service architecture for reliable speech-processing workflows.

PythonFastAPIWhisperStructured API

View the code

Dhaaga - AI Fashion Discovery Platform

Full Stack AI Engineerpersonal

Built and deployed a full-stack AI discovery platform across 100+ brands and 128K+ products using FastAPI, PostgreSQL, pgvector, React, TypeScript, embeddings, semantic search, filtering, NLP intent parsing, and LLMs. Owned data ingestion, search, reranking, fallback, and recommendation workflows while improving product behavior through real search failures and relevance testing.

FastAPIPostgreSQLpgvectorReactTypeScriptEmbeddingsSemantic searchNLPLLMs

Visit the site

Txtura - Generative AI Platform

ML Product Engineerpersonal

Built and shipped a consumer AI product across web and Android using Flutter, FastAPI, Stable Diffusion XL, Hugging Face Diffusers, authentication, cloud storage, and production APIs. Owned the ML-to-product lifecycle including model integration, backend services, GPU optimization, authentication, deployment, failure handling, and production iteration.

FlutterFastAPIStable Diffusion XLHugging Face DiffusersAuthenticationCloud storageNetlifySupabaseFirebaseCloudinaryPollinations API

Visit the site

Translating Words to Pixels: An Effective Text-to-Image System Utilizing Upsampling and CLIP Models

personal

A sole-author research paper published in IEEE Xplore at the International Conference on Electrical, Computer and Energy Technologies (ICECT) 2024.

CLIP

NexGen Tutor: AI-Based Classroom Assistant from PDF Files

personal

A sole-author paper accepted for oral presentation at ACEID 2026 in Tokyo.

Credit-card fraud detection

personal

Trained and compared models including Random Forest and XGBoost on more than 550,000 transactions, handling class-imbalance and using SHAP for feature analysis.

Random ForestXGBoostSHAP

Education

Bachelor of Science

Ghulam Ishaq Khan Institute of Engineering Sciences and Technology (GIKI)Artificial IntelligenceSep 2021 – Jun 2025

Recognition

Patents & publications

NexGen Tutor: AI-Based Classroom Assistant from PDF Files

research paper2026

Sole-author research on a RAG-based classroom assistant using document processing, embeddings, semantic retrieval, and LLM question answering; accepted for oral presentation at ACEID 2026 in Tokyo.

Translating Words to Pixels: An Effective Text-to-Image System Utilizing Upsampling and CLIP Models

publication

Sole-author peer-reviewed research on text-to-image generation using CLIP-based semantic representations and deep-learning image generation, published in IEEE Xplore.

Read it

Social impact & activities

NexGen Tutor: AI-Based Classroom Assistant

ACEID 2026conference talk2026

Accepted for oral presentation in Tokyo.

Hire NuaimaReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more people Create your own Kanvis