Moiz AhmedMansoori
Building AI Systems That Work in the Real World.
// |
Specializing in Agentic Workflows, RAG pipelines, and scalable Machine Learning infrastructure. Building the bridge between foundational models and enterprise solutions.
About Me
As a BS Artificial Intelligence (BSAI) graduate, I specialize in building multi-agent AI systems, RAG pipelines, and full-stack AI applications. During internships at Atlas Battery, CodeCelix, and DUET, I worked on OCR-based document processing, fine-tuned BERT models for NLP tasks, and applied machine learning to EEG signal analysis, gaining experience shipping AI systems that work reliably in production environments.
I build end-to-end AI SaaS products, from data pipelines and model development to APIs, cloud deployment, and LLM-powered workflows. I've independently built and launched AI SaaS applications that automate workflows and solve practical business problems. I learn quickly, adapt to new technologies, and focus on building reliable, maintainable, and scalable AI software.
Experience
IT Intern (Software)
Atlas Battery Limited
- Developed AI-based Warranty Code Detection using EasyOCR and PaddleOCR integrated into the Claim Management System
- Built a LangChain-powered agentic chatbot connected to the company's internal knowledge base
- Built Power BI dashboards to visualize warranty claims and operational data, enabling the IT team to track claim resolution metrics in real time
Machine Learning Engineering Intern
CodeCelix
- Built ensemble models using AdaBoost and XGBoost on balanced and imbalanced datasets
- Handled class imbalance with class weighting and preprocessing techniques
- Fine-tuned BERT-based transformer models for NLP tasks with structured hyperparameter tuning
- Developed end-to-end ML pipelines from preprocessing to validation
Data Analysis Intern
Dawood University of Engineering & Technology
- Analyzed EEG signals for pattern recognition using statistical and ML approaches
- Developed real-time algorithms improving BCI output accuracy
- Assisted in research documentation and collaborated with faculty
Projects
AskSukoon — AI Mental Wellness Companion
- Production-deployed mental wellness SaaS providing 24/7 conversational emotional support and mindfulness guidance.
- Integrated Cognitive Behavioral Therapy (CBT), Acceptance and Commitment (ACT), and DBT therapeutic grounding frameworks.
- Engineered real-time interactive voice streaming pipelines powered by Deepgram for sub-second vocal responses.
- Deployed on Cloudflare Workers edge serverless infrastructure to ensure high availability and low latency.
QueryMind — NL2SQL Agent System
- Built a 10-node LangGraph NL2SQL agent with self-correction (max 3 retries) and pgvector semantic retrieval, achieving 95% pass rate at ~3.1s avg latency.
- Created real-time observability dashboard featuring a live SSE streaming stepper and interactive SVG agent path visualizer.
- Deployed full-stack system: FastAPI backend, Next.js frontend and PostgreSQL handling 1.55M+ database rows.
GIAIC Hackathon II - Spec-Driven Todo
- Shipped a cloud-native software system through a modular 5-phase evolution.
- Built a TDD-tested CLI application transitioning into a full-stack Next.js/FastAPI task platform.
- Configured local Docker and Kubernetes Helm charts for orchestrating multi-container deployments.
- Deployed to Vercel with integrated live user telemetry and performance analytics.
AIRO — AI Research Orchestrator
- Architected a 6-agent LangGraph system automating the full ML pipeline (data ingestion to reports), with a Critic Agent that audits overfitting and data leakage before model selection.
- Logged runs natively to MLflow during parallel ThreadPool training, parsing raw SHAP metrics for dynamic charts.
- Full-stack: FastAPI (Render), Next.js 16/Tailwind v4 (Vercel), Zustand managed SSE logs, and Recharts visuals.
AI Research Paper Analyzer
- Engineered custom RAG pipeline using PyMuPDF to extract layout-aware summaries and citations.
- Built Flask REST API backend with real-time, stage-based progress polling for frontend.
- Implemented FAISS vector database and MMR search for semantic, grounded document retrieval.
Brain Tumor Detection
- Fine-tuned a VGG-16 CNN model reaching 94% classification accuracy on 7,200 MRI images.
- Evaluated models across 4 classes: glioma, meningioma, pituitary, and normal brain scans.
- Designed OpenCV preprocessing stages to improve contrast and isolate region highlights.
- Implemented Grad-CAM visualization to render heatmaps of the network's visual attention.
Technical Capabilities
AI & Machine Learning
LLMs & Agentic AI
Data Science
Tools & Infrastructure
Education
Bachelor of Science in Artificial Intelligence
GraduatedDawood University of Engineering and Technology
Karachi, Pakistan
September 2022 - June 2026
Certifications
Certified Cloud Applied Generative AI Engineer (GenEng)
Governor House — GIAIC
Feb 2023 – Present
Get in Touch
I'm always open to discussing AI engineering opportunities, collaborations, or just geeking out about machine learning. Feel free to reach out!