Muhammad Mohid Subhani

Muhammad Mohid Subhani

AI/ML Engineer specializing in RAG Pipelines and Agentic Systems

Based in: Lahore, Punjab, PakistanCurrently: Software Engineer AI/ML, DevsincExperience: 2 years

Hire me

A Computer Science graduate and Software Engineer with extensive experience in building Retrieval-Augmented Generation (RAG) pipelines and fine-tuning large language models. He has a proven track record of developing AI research assistants and agentic systems across the software and gaming industries.

Experience

Software Engineer AI/ML

DevsincworkApr 2025 – Present

Summary

Built a complaint mapping system to link customer queries with relevant employees using NLP techniques. Worked on achieving reproducible results using Retrieval-Augmented Generation (RAG) pipelines. Developed an AI Research Assistant that retrieves citations from research papers using a domain-specific embedder, SciBERT, improving retrieval quality by 15%. Created a social media and influencer marketing analytics tool to track posts, likes, comments, and engagement using official platform APIs. Utilized GPT-4.1 to classify influencer comments into positive, negative, and neutral categories for sentiment tracking.

Results

  • Improved retrieval quality by 15% for the AI Research Assistant using SciBERT.
NLPRAGSciBERTGPT-4.1APIs

AI Engineer

Creative DistrictcontractApr 2025 – Present

Summary

Built an Agentic AI System to assist mobile game companies in assisting them through their end-to-end project. Created FTUE (First-Time User Experience) and GDD (Game Design Document) from gameplay videos, turned them into frames, and used Gemini 2.5 Pro to create game documents. Integrated Firebase Analytics and Google Ads to track user gameplay data for performance analysis. Fine-tuned GPT-4o on large-scale game datasets to generate personalized Retention, Monetization, ROAS, etc suggestions. Implemented a RAG system to retrieve GDDs of similar games, enhancing model recommendations with contextual knowledge.

Results

  • Improved recommendation quality by around 18% through fine-tuning GPT-4o on game-level examples.
  • Achieved roughly a 10% improvement in retention-related KPIs in experiments.
  • Fine-tuned GPT-4o on roughly 8,000 game-level examples combining player behavior, retention metrics, and historical optimization recommendations.
Agentic AIGemini 2.5 ProFirebase AnalyticsGoogle AdsGPT-4oRAGLangGraphLangSmith

Machine learning Engineer Intern

NetSol Technologiesinternship2025 – Mar 2025

Summary

Gained knowledge in NLP (Natural Language Processing), exploring techniques for text analysis and understanding language patterns. Integrated GROG API key to retrieve responses from LLMs, enabling efficient problem-solving. Utilized Unsloth for model quantization, optimizing performance and reducing model size for efficient deployment, while applying LoRA (Low-Rank Adaptation) and pruning techniques to fine-tune and compress the LLaMA model for improved efficiency and reduced computational overhead. Evaluated an LLM using perplexity, HumanEval, and HellaSwag benchmarks to assess performance, coherence, and reasoning capabilities. Developed a RAG evaluation pipeline to systematically measure retrieval-augmented generation performance, ensuring accurate and contextually relevant responses.

What I did

  • Used 4-bit quantization and LoRA to manage model size and fine-tuning footprint

Results

  • Reduced LLaMA model size by roughly 50–60% through 4-bit quantization
  • Improved inference latency by around 30–40% in tests
NLPGROG APILLMsUnslothLoRAPruningLLaMAHumanEvalHellaSwagRAG4-bit quantization

Skills

Technical

PythonPython
Scikit Learn
RAG
Fine-Tuning
Generative AI
Natural Language Processing
Pandas
NumPy
TensorFlowTensorFlow
PyTorchPyTorch
LangChain
LlamaIndex
ChromaDB
GitGit
GithubGithub
LangGraph
LangSmith
Seaborn
Matplotlib
MS SQLMS SQL
PostgreSQLPostgreSQL
Overleaf
Azure (ACR, AKS)
AWS (ECR, EKS, EC2, S3)
RR
Unsloth
TypeScriptTypeScript
GPT-4o
4-bit quantization
ReactReact
LoRA

Areas of expertise

Quick Learning
Analytical Mindset

Projects

GameLyft

personal
Used LangGraph to orchestrate a workflow retrieving game and player analytics, pulling historical recommendations via RAG, and generating and validating improvement ideas.
LangGraphLangSmithRAG

AI analytics platform

personal
Owned a feature end-to-end, building the React frontend and integrating it with Python APIs while handling data flow, error states, and the ML/RAG pipeline connection.
ReactPythonML/RAG pipeline

Education

BS Computer Science

University of Engineering and Technology, LahoreComputer ScienceOct 2020 – Jun 2024

Recognition

Certifications

Machine Learning Specialization

Stanford and DeepLearning.AI2024

Introduction to AWS

AWS Academy2023

Introduction to Business

LUMS2022
Hire MuhammadReach out about a role, a contract or a conversation.For recruiters

Muhammad's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis