Muhammad Ali Nasir / ML Engineer
I build the parts of AI that still work after the demo ends.
Agentic systems, retrieval pipelines, and production ML — presented with the decisions and evidence behind the build.
I make complex systems
easier to inspect.
easier to inspect.
07systems shipped
01publication under review
01open-source PyPI package
∞questions welcome
01 / Selected work
Systems built for real constraints.
From causal pricing to self-hosted inference, these projects are about making intelligent systems useful outside the notebook.
02 / Research
LightUHope
First author · Under review at Springer
A lightweight transformer for four-class Urdu hope-speech classification, built around a 24,124-sample dataset and interpretable feature analysis.
0.92macro F1
3.2Mparameters
97%fewer than mBERT
03 / Stack
Tools chosen for the problem.
Agentic AI
LangGraph · CrewAI · LangChain · LlamaIndex · MCP
ML / DL
PyTorch · scikit-learn · Transformers · PEFT · LoRA / QLoRA
Retrieval
Neo4j · Qdrant · ChromaDB · LanceDB
Backend / MLOps
FastAPI · Django REST · Docker · PostgreSQL · llama.cpp · SSE
04 / Education
Training behind the systems.
BS Computer and Information Sciences
Pakistan Institute of Engineering and Applied Sciences (PIEAS), Islamabad
3.44/4.0CGPA
96.7thNSCT percentile
2026graduated
Certifications
01ML for Production
(MLOps)DeepLearning.AI
02NLP
SpecializationDeepLearning.AI
03MLOps FundamentalsDataCamp
04Agentic AI with CrewAI &
LangGraph2025
05 / Contact