Mobin Shahidi — AI Engineer & Software Developer

Career Highlights

My journey began in full-stack software engineering, where I co-founded a startup and built a production SaaS platform from the ground up. I designed system architecture, implemented 33 REST APIs, structured relational databases, and deployed across multiple production releases — serving over 100 users in real-world environments.

Driven by a deep interest in intelligent systems, I transitioned into AI and recommender systems research, focusing on information quality and trust in algorithmic newsfeeds. At the University of Potsdam, I engineered 41 large-scale information-quality signals from 5M+ tweets and designed a quality-aware recommender system that improved average recommended source quality by +3 points while maintaining precision.

Alongside this, I developed a deep learning explainable recommendation framework combining NLP, multi-task learning, and neural text generation over 151K reviews. The system achieved strong predictive performance (MSE 0.992) and produced personalized explanations with high semantic consistency (~0.97 similarity), bridging research innovation with practical evaluation.

Today, my focus lies at the intersection of production engineering and applied AI. I deploy ML systems using Docker and AWS (SageMaker), integrate observability and monitoring, build FastAPI-based inference services, and design workflow-aware LLM systems using LangChain — combining research rigor with scalable software delivery.

Skills & Expertise

🤖 AI & Machine Learning

Recommender Systems & Ranking
LLM Integration (LangChain, RAG, Fine-tuning)
Explainable AI (XAI)
Deep Learning (PyTorch, TensorFlow)
Model Evaluation & Experiment Design

⚙️ Backend Engineering

FastAPI & REST API Design
ASP.NET Core & C#
System Architecture & Modular Design
SQL Databases & Schema Design
Data Pipelines & ETL Workflows

☁️ MLOps & Infrastructure

Dockerized Model Deployment
AWS (SageMaker, CloudWatch)
CI/CD Pipelines
Model Observability & Logging
Git & Version Control

📊 Product & Research

Experimental Design & Statistical Analysis
Feature Engineering at Scale
Stakeholder Communication
Cross-functional Collaboration
Scientific Writing & Publications

Projects

Selected research and production projects in AI, recommender systems, and scalable web engineering.

Featured Projects

📰 Quality-Aware News Source Recommender

Designed and deployed a quality-aware recommender system that integrates 41 engineered information-quality signals derived from 5M+ tweets to improve news source reliability.

Recommender Systems ML Pipelines AWS Docker

Impact: Increased average recommended source quality by +3 points while maintaining baseline precision.

Stack: Python · Pandas · Scikit-learn · Docker · AWS · PyTest

📸 PhotoCans – Image Management Platform (Startup)

Co-founded and delivered a commercial full-stack web platform serving 100+ users with licensing, product management, and operational dashboards.

ASP.NET Core REST APIs SQL Server IIS

Impact: Implemented 33 REST endpoints and 12+ business workflows; deployed and maintained across multiple production releases.

Stack: C# · ASP.NET Core · JavaScript · SQL Server · IIS

📊 News Source Quality Predictor (Production ML)

Built and deployed a regression model estimating news source quality using network and diversity features, exposed via API and interactive UI.

FastAPI SageMaker Model Deployment Observability

Impact: Serverless SageMaker deployment with logging and CloudWatch monitoring for production inference.

Stack: Python · FastAPI · AWS SageMaker · CloudWatch · Docker

Research & AI Systems

🔍 Exogenous Cues to Information Quality

Engineered contextual quality variables to enhance algorithmic newsfeeds and improve information reliability in recommender systems.

Feature Engineering Evaluation Recommender Systems

Stack: Python · Scikit-learn · Statistical Modeling

💡 Deep Learning Explainable Recommender

Built a large-scale NLP + recommendation framework over 151K reviews combining multi-task learning and neural text generation to produce personalized recommendations and explanations.

PyTorch TensorFlow NLP Multi-task Learning

Impact: Achieved MSE 0.992 in rating prediction and ~0.97 semantic similarity in explanation generation.

Stack: Python · PyTorch · TensorFlow · RNN · CNN · Transformers · GPU Training

📡 Automated Reddit Scraper

Developed a scalable Reddit data collection pipeline extracting threads, comments, and interaction metadata for political discourse research.

Data Pipeline APIs NLP Data

Stack: R · Python · API Integration · Data Validation

Software & Product Engineering

🤖 Workflow-Aware AI Assistant

Built a state-aware AI assistant integrated into a Next.js / Blazor application, orchestrating OpenAI APIs and local LLMs (Mistral) using LangChain for structured task execution.

LLMs LangChain Next.js System Orchestration

Stack: FastAPI · LangChain · OpenAI · Mistral · Next.js