Sanwal Bilal

Sanwal Bilal

AI Engineer & Architect | Specialist in Computer Vision, RAG, and Autonomous Agents

Based in: Islamabad, PakistanMost recently: Junior Machine Learning Engineer, AI Soft (Pvt) LtdExperience: 4 years

Hire me

Junior Machine Learning Engineer with a proven track record of architecting end-to-end AI pipelines. Expert in Computer Vision (YOLO, InsightFace) and LLM orchestration (LangGraph, RAG). Developed production-ready systems for real-time safety monitoring and autonomous reconnaissance. Passionate about bridging the gap between complex AI research and practical, scalable hardware/software integrations.

Experience

Junior Machine Learning Engineer

AI Soft (Pvt) LtdIslamabad, PakistanworkFeb 2023

Summary

Developed, deployed, and maintained scalable AI/ML applications using Python, TensorFlow, and PyTorch, improving system performance by 30% and reducing processing time by 25%. Designed and deployed ML models for AI-driven features, achieving 90%+ accuracy across key use cases and enhancing user engagement by 20%. Integrated third-party AI/ML APIs into existing web applications, reducing development time by 40% and improving feature delivery speed. Implemented data-driven optimization strategies, increasing model efficiency by 25% and reducing error rates by 15%. Built and optimized NLP pipelines using open-source LLMs for summarization and information extraction, improving response relevance by 35%. Led NLP and computer vision initiatives, improving team productivity by 20% through collaboration and mentorship.

What I did

  • Built and optimized NLP pipelines using LangChain and model pruning techniques for high-speed summarization and information extraction

Results

  • Architected and deployed real-time inference workflows using AWS Lambda and Docker, reducing latency for high-speed detection tasks.
  • Improved system performance by 30% and reduced processing time by 25% through serverless optimization.
  • Achieved 90%+ accuracy across key use cases and enhanced user engagement by 20%.
  • Reduced development time by 40% and improved feature delivery speed.
  • Increased model efficiency by 25% and reduced error rates by 15%.
  • Improved response relevance by 35%.
  • Improved team productivity by 20% through collaboration and mentorship.
PythonTensorFlowPyTorchNLPLLMsComputer VisionAWS LambdaDockerServerless ArchitecturesDatabricksLangChainModel Pruning

Skills

Technical

Deep Learning
Python (3.x)Python (3.x)
LLMs
Generative AI
PyTorchPyTorch
NLP
Computer Vision
TensorFlowTensorFlow
Ethical Hacking
OOP
Data Structures & Algorithms
Transformer Models
Model Evaluation
NumPy
Pandas
Scikit-learn
Keras
Hugging Face Transformers
REST API Development
AWS
SQLSQL
Relational Databases
GitGit
LinuxLinux
GCPGCP
Databricks
DockerDocker
Serverless Architectures
Data Visualization
System Monitoring & Alerting
SQLAlchemy (ORM)SQLAlchemy (ORM)
Distributed & Eventually Consistent Systems
Microservices
Design Patterns
Message Brokers
Event-Driven Systems
FlaskFlask
DjangoDjango
Dependency Injection
SOLID
Azure

Projects

AI-Powered PPE Detection & Reporting System

Architectpersonal
Architected a scalable AI pipeline combining YOLO-based computer vision and LLM-based reporting for intelligent safety monitoring, specifically optimized for detecting safety shoes on fast-moving personnel. Engineered real-time inference workflows with optimized frame processing and detection pipelines. Developed automated incident reporting using LLMs, reducing manual reporting effort significantly. Designed modular micro-pipeline enabling seamless integration with edge hardware and cloud systems.
Computer VisionLLMYOLOPythonAWS LambdaDockerRaspberry Pi
View the code

Autonomous AI Reconnaissance System

Architectpersonal
Architected an LLM-powered autonomous agent system for OSINT-driven reconnaissance and intelligence gathering. Engineered dynamic tool orchestration using LangGraph, enabling context-aware decision loops. Built scalable multi-tool integration pipeline for real-time data enrichment and structured insight generation. Designed extensible framework supporting cybersecurity research and automated intelligence workflows.
LLMsLangGraphOSINTCybersecurityEthical Hacking Tools
View the code

Real-Time Missing Person Detection with RAG-Based Alerting

Developerpersonal
Built end-to-end pipeline: Video Stream → Face Detection → Embedding Generation → Vector Search → Identity Matching. Leveraged InsightFace (buffalo l) for high-accuracy facial embeddings; implemented RAG pipeline for large-scale record matching. Designed automated alert system with multi-channel notifications (SMS & Email) for real-time incident reporting. Achieved ~90% recognition accuracy with minimized false accept/reject rates and near real-time detection-to-alert latency.
Face RecognitionVector SearchInsightFaceRAGSMSEmailDatabricks
View the code

Education

B.S. in Computer Science

University of MianwaliComputer ScienceGPA 3.38Apr 2026

Recognition

Certifications

AI and Cybersecurity (Hacking by Gen AI)

Google & HEC Pakistan
Hire SanwalReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis