Saqib Iqbal

AI engineer with 4 years building production LLM and computer-vision systems. Recent work: RAG pipelines on Pinecone, autonomous agent evaluation in LangGraph, fine-tuning on Azure ML, and low-latency FastAPI inference services. Published research on hybrid CNN-Transformer deepfake detection.

Founder & Lead ML Engineer at Orbimatrix (2024-04 – 2026-01)

Self-founded ML & Web3 consultancy

• Designed evaluation harnesses for autonomous LLM agents in LangChain and LangGraph, measuring tool-use accuracy and reasoning-chain quality across GPT-4o and Claude 3.5 deployments.
• Owned full ML lifecycle on Azure ML, data curation, feature engineering, fine-tuning with DVC and MLflow for experiment tracking and reproducible model versioning.
• Built high-concurrency RAG pipelines on Pinecone backed by Polars and DuckDB; served models through BentoML with drift monitoring and automated feedback loops.
• Delivered Web3 integration work alongside ML for client products: token gating, on-chain triggers, and smart-contract data feeding model pipelines.
• Built an end-to-end industrial monitoring system with PostgreSQL, Python ETL, Power BI KPI dashboards, and anomaly detection for predictive maintenance.

CTO (Contract) at NIBLS (2023-10 – 2024-04)

• Set technical direction for a Smart Ball IoT product combining sensor telemetry with on-chain triggers via Chainlink CCIP.
• Built async telemetry ingestion in Python (asyncio, multithreading) handling thousands of concurrent sensor heartbeats from sports IoT devices.
• Deployed LangGraph agent workflows that analyze player-performance data and generate GPT-4o executive reports; served LLM inference on Ray Serve behind FastAPI microservices on GCP Cloud Run.
• Built a Germany tech job analytics platform that extracts role requirements from job descriptions and visualizes skill demand across AI, data, backend, and robotics roles.

AI Engineer (Contract) at Churn (2023-11 – 2024-01)

• Trained a credit-card fraud detection model on a 500k+ transaction dataset, using SHAP feature attribution and time-series cross-validation to eliminate target leakage.
• Built ingestion services pulling transactional data from fintech APIs into SQL with end-to-end data-quality checks; delivered KPI dashboards in Power BI and Seaborn for fraud and segmentation reporting.
• Built a ROS2 warehouse delivery robot simulation with navigation, obstacle avoidance, task assignment, and performance logging in Gazebo/RViz.
• Created a computer vision quality inspection system using YOLO/OpenCV to detect product defects and serve results through a FastAPI web interface.

B.Sc. in Computer Science – University of Engineering & Technology (UET)