Hi DEV 👋
I’m Muhammad Ibrahim Hashmi, a final-year BS Artificial Intelligence student from Islamabad, Pakistan.
Most of my work so far has been around Machine Learning, Computer Vision, Python, FastAPI, and building AI applications end to end.
I joined DEV because I want to document something I’ve been realizing more and more:
Training a model is only one small part of building a useful AI system.
Moving beyond the model
When I first started working with AI, most of my attention naturally went toward things like:
- model selection
- training
- accuracy
- computer vision pipelines
- predictions
But as I started building complete applications, the questions changed.
How does the model receive data?
How do you expose inference through an API?
What happens when the input is invalid?
How do you connect the model to a frontend?
How do you make the system reliable enough for someone other than you to use?
That is where tools like FastAPI, REST APIs, testing, deployment, and software-engineering practices started becoming much more important to me.
Some things I’ve been building
A few projects that shaped this thinking:
UniTime-AI
An AI-assisted university timetable application where machine learning is part of a larger product workflow rather than the entire product.
Building it forced me to think about how AI decisions fit inside an application and how software architecture affects the usefulness of the model.
Pakistani Currency Assistant
A computer-vision application for recognizing Pakistani currency.
One of the most useful lessons from this project came from failure cases.
A model can produce a confident prediction and still be completely wrong.
That sounds obvious, but seeing a real false-positive case changes how you think about production AI.
A confidence score is not the same thing as reliability.
It made me think much more seriously about:
- out-of-distribution inputs
- false positives
- confidence thresholds
- validation
- testing with unexpected inputs
- safety gates around model predictions
AI Resume Studio
A Python and Streamlit application for creating and analyzing resumes, including ATS-style analysis and resume-to-job-description matching.
This project pushed me toward thinking about AI as part of a workflow instead of as an isolated prediction.
The useful question becomes:
What decision or task is this AI actually helping someone complete?
What I’m focusing on now
Right now I’m especially interested in the engineering side of AI:
- Computer Vision
- Python
- FastAPI
- REST API design
- model integration
- reliable inference pipelines
- deployment
- testing
- building AI systems that solve complete problems
I’m also working on Pur-Aman Pakistan, my final-year project exploring multimodal AI for public-safety incident reporting.
Why I’m writing here
I don’t want this profile to become a collection of AI-generated tutorials or posts about technologies I haven’t actually used.
My plan is simpler:
Build → encounter a real problem → understand it → write about what I learned.
So future posts will probably cover things like:
- mistakes I make while building AI systems
- Computer Vision failure cases
- FastAPI and ML integration
- turning ML experiments into usable applications
- lessons from real project architecture
- reliability problems that accuracy metrics don't reveal
I’m still early in my career, which is exactly why I think documenting the process will be useful.
There will be things I get wrong.
There will be things I understand differently six months from now.
That is part of the point.
If you work in AI, ML, Computer Vision, Python, or production AI systems, I’d be happy to connect and learn from your experience.
Thanks for reading my first post on DEV. 🚀
Top comments (0)