Skip to content

Shahid Ali / Applied AI Engineer

Building intelligent systems, from research to real-world applications.

I'm an Applied AI Engineer with a Master's degree in Artificial Intelligence from the University of Verona, Italy. I work on AI agents, machine learning, deep learning and reinforcement learning, with a focus on building reliable applications and testing how well they work.

Open to AI engineering, machine learning and research opportunities.

Download CVGet in touch
Now
Freelance AI engineer · Appycodes
Education
MSc Artificial Intelligence · University of Verona
Based in
Verona, Italy
Focus
Agentic AI · ML · DL · RL
Looking for
AI engineering roles and research opportunities
MSc AI
University of Verona, 2026
70.1%
best RL agent vs baseline in my thesis research
75% vs 0%
RECALL-Agent with and without memory (offline ablation)
UK · South Africa
freelance client projects

about

About me

I'm an Applied AI Engineer based in Verona, Italy, with a Master's degree in Artificial Intelligence from the University of Verona.

My work focuses on intelligent agents, machine learning and deep learning applications, and reinforcement learning systems. The part I find most interesting is making AI applications reliable: integrating tools, managing data, evaluating models and improving them.

My thesis was on multi-agent reinforcement learning for building energy management. Alongside it I've built RAG applications, AI agents and computer vision models. I also freelance with Appycodes, building LLM applications and web platforms for clients, most recently a lettings CRM for a UK agency.

I'm now looking for a place where I can contribute to real AI systems, keep learning from experienced engineers and researchers, and take on meaningful technical problems.

  • Start from a baseline

    I like a clear question and something to compare against, run over several seeds, so I can tell whether a change actually helps.

  • Finish the last mile

    Clean code, reproducible runs and a working interface, not just a result in a notebook.

  • Research and engineering

    Comfortable reading papers and running experiments, and equally at home building the FastAPI and Next.js apps around them.

  • Work with real clients

    Freelancing taught me to scope work with non-technical people, agree what done means and hand over cleanly.

research

What I work on

Three areas I keep coming back to, in research and in my own projects.

Agentic AI and LLM systems

  • AI agents and multi-tool workflows
  • LangGraph, LangChain, LlamaIndex and CrewAI
  • Tool calling and orchestration
  • Memory and state management
  • RAG and document question answering

Machine learning and deep learning

  • Model training and evaluation
  • CNNs, Transformers and LSTMs
  • PyTorch and TensorFlow
  • Computer vision and explainability
  • Data preprocessing and evaluation pipelines

Reinforcement learning

  • SAC and PPO with Stable Baselines3
  • Multi-agent RL: MAPPO and MADDPG
  • CityLearn energy management environments
  • Reward design, including an LLM-assisted loop
  • Evaluation against baselines across seeds

MSc thesis and research internship

Multi-agent reinforcement learning for building energy management

University of Verona · supervised by Prof. Alberto Castellini · CityLearn 2023

I built five smart-building energy environments, each harder than the last, moving from single-building control to multi-building coordination with PV, battery storage, HVAC and domestic hot water. I compared SAC, PPO, MADDPG and MAPPO over three seeds against a random baseline, and added an LLM loop to help design the reward.

70.1%
improvement over the baseline with the best agent
219 → 0
demand peaks over 720 control steps with the best agent

projects

Things I've built

Three projects I'd show first, each with a longer write-up, then smaller ones. Source code is linked where the repository is public.

01/Agents · memory

RECALL-Agent

Read the case study Source code
What it does
A coding agent that learns from its mistakes. Each run can leave a short, reusable lesson that later tasks retrieve.
What I built
Planner, worker and critic agents in LangGraph, a verifier that runs every solution against a check() function in a time-limited subprocess, and a lesson memory in SQLite and ChromaDB with scoring and pruning.
Result
75% success with memory and 0% without, in an offline ablation over 9 task families, 4 rounds and 3 seeds. In that setup memory is the only way to pass, so it isolates the memory loop. A smaller live run with an OpenAI model went from 48% to 54%.
Two charts comparing memory on and memory off. Cumulative success rises to 0.75 with memory on and stays at 0 with memory off; per-round success is 0 in round 1 and 1.0 in rounds 2 to 4 with memory on.
Learning curve from the offline benchmark, from the project repository.

LangGraph · ChromaDB · SQLite · FastAPI · OpenAI and Anthropic APIs

02/Agents · RL

RL Environment Auto-Designer

Read the case study Source code
What it does
Five cooperating LLM agents turn a plain-English problem into a working simulation, fix their own code, train a policy and write a report.
What I built
A Designer that writes a typed environment spec, a Coder that turns it into Gymnasium code, an Executor that runs it in a subprocess and sends errors back for repair, a Trainer and a Reporter, all behind FastAPI with live progress in Next.js.
Result
Each run produces a trained SAC or PPO policy, a reward curve, a results CSV and a written report explaining the design.

FastAPI · Gymnasium · Stable Baselines3 · Next.js · Claude and GPT APIs

03/Voice · automation

AI Voice Receptionist

Read the case study Source code
What it does
Answers calls, picks up intent and urgency, saves a scored lead with a summary and hands off to a person when needed.
What I built
The voice flow (browser Web Speech API or real-time LiveKit), lead scoring and handoff logic, a FastAPI backend on SQLite and a Next.js dashboard. It runs on OpenAI, a local Ollama model or a rule-based fallback.
Result
The whole flow runs end to end without API keys, and a pytest suite covers the API.
Lead detail page in the receptionist dashboard showing a hot lead for personal training: contact details, urgency high, lead score 10 out of 10, a conversation summary, a recommended next action and a suggested follow-up message.
Lead detail screen from the dashboard, filled with demo data.

LiveKit · OpenAI · FastAPI · Next.js · SQLite

more projects

RAG

Document Assistant

Answers questions about uploaded PDFs with file and page citations, and can summarise and compare documents.

LangChain · LlamaIndex · ChromaDB · FastAPI

Source code

Deep learning · XAI

Lung Cancer Image Classification

A CNN that reached 99.8% accuracy on the LC25000 dataset with 5-fold cross-validation. I implemented the explainability methods from scratch.

PyTorch · Captum

Agent · ML

AI Data Analyst Agent

Upload a CSV or Excel file and ask questions in plain language. It suggests charts, flags anomalies, trains simple models and exports a report.

LangGraph · pandas · scikit-learn · Plotly

Source code

Agent

Multi-Tool Research Agent

Researches a topic across web search, arXiv and Wikipedia, rates sources by reliability and writes structured briefs.

LangGraph · LangSmith · Tool calling

Source code

Multi-agent

CrewAI Business Agent System

An operations workspace for small businesses: a protected Next.js dashboard over FastAPI and CrewAI, with persistent retrieval in ChromaDB.

CrewAI · FastAPI · Next.js · ChromaDB

Source code

Time series

NASA Engine RUL Prediction

Remaining-useful-life prediction for turbofan engines on NASA C-MAPSS with LSTM and BiLSTM models and an engine-aware train/validation split.

LSTM · TensorFlow · C-MAPSS

Source code

experience

Experience and education

Experience

  1. Ongoingnow

    Freelance AI Engineer and Full-Stack Developer

    Appycodes · remote

    Building LLM applications and web platforms for international clients, from requirements to handover. Currently a lettings CRM for a UK agency (Next.js, Supabase); delivered a website for a client in South Africa.

  2. May 2026 – Oct 2026

    Research Intern, Reinforcement Learning

    University of Verona

    Built five smart-building energy environments and compared SAC, PPO, MADDPG and MAPPO over three seeds. The best agent beat the baseline by 70.1% and cut demand peaks from 219 to 0. Added an LLM reward-design loop.

  3. Mar 2026 – May 2026

    Machine Learning Engineer Intern

    University of Verona

    Classification and regression models in scikit-learn, from cleaning raw data with pandas and NumPy to training and validating linear models, random forests and SVMs.

  4. Dec 2024 – Feb 2025

    Deep Learning Intern

    University of Verona

    CNN image classifiers in PyTorch, evaluated with accuracy, precision, recall and F1, and integrated into application pipelines with the team.

Education

  1. 2023 – 2026

    MSc in Artificial Intelligence

    University of Verona, Italy

    Completed October 2026. Thesis on multi-agent reinforcement learning for building energy management, supervised by Prof. Alberto Castellini.

  2. 2018 – 2022

    Bachelor of Engineering

    Mehran University of Engineering & Technology (MUET), Pakistan

  3. Completed

    Agentic AI course (one year)

    Krish Naik Academy

skills

Tools I use

The same list as my CV, grouped by where I use it.

GenAI and agents

  • LLMs
  • RAG
  • AI agents
  • Prompt engineering
  • LangChain
  • LangGraph
  • LlamaIndex
  • CrewAI

Machine learning and deep learning

  • PyTorch
  • TensorFlow
  • scikit-learn
  • Hugging Face
  • CNNs
  • Transformers
  • LSTMs
  • Computer vision

Reinforcement learning

  • SAC
  • PPO
  • MAPPO
  • MADDPG
  • Stable Baselines3
  • Gymnasium

Engineering

  • Python
  • SQL
  • JavaScript
  • FastAPI
  • Next.js
  • React
  • Supabase
  • ChromaDB
  • PySpark
  • Git

Languages: English (B2), Italian (A2), Urdu and Sindhi (native).

contact

Get in touch

I'm open to opportunities in Agentic AI, Machine Learning, Deep Learning, Reinforcement Learning and applied AI research, including PhD positions. I'd like to work with a team building practical AI systems. Email is the quickest way to reach me.

shahidali63331@gmail.com
LinkedIn
in/shahid-ali-23250a17b
GitHub
Shahidali78
Location
Verona, Italy