CV
Open to work · Islamabad

Hi there, I'm

Nimra Faryal

I build machine learning models

AI / ML Engineer and Computer Science graduate from UET Taxila. I train models on real-world data and build the systems around them: detection and tracking pipelines, LLM apps, APIs and sensor networks.

View my work Download CV
Daily drivers
CGPA / 4.03.69
AI projects10+
4th semester rank1st
Hackathons2
PyTorch Scikit-Learn Hugging Face Transformers Neural Networks CNNs Transfer Learning NLP Random Forest Feature Engineering Model Evaluation GPU Training Llama 3.3 OpenAI API Groq API DeepSeek V3 Whisper BERT Prompt Engineering RAG Vector Search Qdrant AI Agents Sentiment Analysis ElevenLabs YOLOv8 YOLOv11 YOLO11-OBB YOLO11-Pose Object Detection Object Tracking ByteTrack Re-ID Instance Segmentation Pose Estimation Homography OpenCV MediaPipe Roboflow Pandas NumPy Matplotlib Seaborn Jupyter EDA Data Cleaning Preprocessing Regression Classification Correlation Analysis FastAPI Flask REST APIs JSON Pytest Streamlit Automated Data Pipelines Python C++ SQL PostgreSQL SQLite Raspberry Pi Arduino Sensor Integration Wireless Sensor Networks IoT Systems Real-Time Data Collection Git GitHub Google Colab Kaggle Anaconda VS Code
[01]About

I train models, then build what runs them.

README.md

# What I work on

I'm fascinated by what happens when code meets the real world: data starts showing patterns, a model learns to see, and an idea becomes something that actually runs. That curiosity has pulled me across machine learning, deep learning, computer vision, NLP and IoT. I've forecast Walmart sales with a Random Forest, and trained YOLO models that track basketball players, catch shoplifters and spot predators on a farm.

I like looking at a problem from a few angles and bringing different tools together to solve it, whether that's an AI interviewer built on Llama 3.3 and Whisper, a FastAPI security gateway for AI agents, or Raspberry Pi sensor nodes for my final year project. For me, engineering is more than getting a model to train. It starts with understanding the problem and ends with something people can use. I'm always learning, and usually halfway through building something new.

Islamabad UET Taxila '26Python · PyTorch · FastAPI open to work
~/nimra — zsh

› whoami

nimra-faryal · ai / ml engineer

› cat focus.txt

ml · deep learning · computer vision · llm apps · iot

› ls ./stack

python pytorch yolo opencv transformers fastapi postgresql

› pytest action-gate/tests -q

52 passed

›

Projects shipped
138 computer vision · 4 NLP & LLM · plus ML, backend and IoT
/01

Machine Learning

Cleaning, EDA, feature engineering and model evaluation on structured data.

/02

Deep Learning & Vision

Custom YOLO training, tracking, pose and segmentation on real footage.

/03

NLP & LLMs

LLM apps with Groq, Whisper, Transformers, RAG and vector search.

/04

Backend & IoT

FastAPI and Flask APIs, PostgreSQL, Raspberry Pi and Arduino sensors.

[02]Experience

git log --career

Hackathons, my final year project and leadership, newest first.

  1. Oct 12 – 18, 2026HEAD → main

    AMD Developer Hackathon: ACT III

    @ ACTION GATE: AI Agent Security Gateway · lablab.ai · Team Veylora

    feat/ai-agent-security
    • +Built a security gateway that decides whether an AI agent's actions are allowed, need approval or are blocked
    • +FastAPI backend with DeepSeek V3 on AMD Developer Cloud, Streamlit dashboard, 52 automated tests
    FastAPI DeepSeek V3 AMD Developer Cloud PytestProject page Certificate
  2. Nov 14 – 19, 2025

    AI Genesis Hackathon

    @ Multimodal Insurance Claims Processor · lablab.ai · Team Stellar Engineers

    feat/vector-search
    Qdrant BERT OpenAI Streamlit
  3. Final year

    Final Year Project: IoT Forest Monitoring Network

    @ UET Taxila · team of 3

    feat/iot-sensor-network
    Raspberry Pi Arduino Flask PostgreSQL
  4. 2026

    Aspire Leaders Program

    @ Cohort 4

    feat/leadership
[03]Stack

The toolbox.

Languages, frameworks and tools I have used in real projects, from data cleaning to hardware.

~/skills8 folders · 76 files

AI / Machine Learning

~/skills/ai/

PyTorch
Scikit-Learn
Hugging Face
Transformers
Neural Networks
CNNs
Transfer Learning
NLP
Random Forest
Feature Engineering
Model Evaluation
GPU Training
[04]Work

Selected projects.

Machine learning, deep learning, computer vision, LLM apps, backend and IoT. Every vision clip is real model output.

0.9salert after grab
17pose keypoints
~5FPS on CPU

featuredComputer Vision · Featured on LinkedIn

CCTV Suspicious Behaviour Detection

Tracks every customer on store CCTV and flags a shelf grab followed by a hand moving to the pocket, with a live alert dashboard and risk score per person.

  • Pose-keypoint behaviour rules with zone polygons over shelves
  • Risk score per tracked ID, live alerts and suspect snapshot
  • Alert fired 0.9 s after the grab on the test clip
// pipeline
YOLO11-Pose→ByteTrack→Zone rules→Risk score→Alert
Python YOLO11-Pose ByteTrack OpenCV
LinkedIn demo

› ls ./projects --filter=all

~/projects/walmart
Predicted vs actual sales scatter plot with RMSE 15,230 and R² 0.945

Machine Learning

Walmart Weekly Sales Forecasting

Forecasts weekly Walmart sales from historical sales, store details and external factors such as markdowns, temperature, CPI and unemployment.

  • Cleaned and merged store, sales and economic datasets
  • EDA, correlation analysis and feature importance
  • Random Forest regressor: R² = 0.945, RMSE = 15,230
GitHub
~/projects/interview
// pipeline
  1. 1Role & CV
  2. 2Llama 3.3 writes questions
  3. 3Whisper transcribes answer
  4. 4LLM scores answer
  5. 5Proctor: gaze + phone
  6. 6Report

LLM Application + Computer Vision

AI Interview Studio

A complete AI interview system on one laptop: an LLM writes role-specific questions and scores the answers, while live proctoring tracks gaze and flags phones.

  • Groq Llama 3.3 70B generates questions and evaluates answers
  • Whisper turns spoken answers into text
  • MediaPipe gaze tracking and YOLOv8n phone detection give a 0–100 integrity score
+1
GitHub
~/projects/sentiment
// pipeline
  1. 1Your text
  2. 2Transformers pipeline
  3. 3Label + confidence

NLP

Sentiment Analysis Web App

A Streamlit app that classifies any text as positive or negative using a Hugging Face Transformers sentiment model.

  • Positive / negative classification
  • Shows the model's confidence score
  • Simple interactive interface
GitHub
~/projects/farm

Computer Vision

Farm Predator Detection

Detects predators such as a mongoose on a poultry farm camera, tracks every chicken with its own ID and raises an intrusion warning.

  • Custom dataset created and annotated in Roboflow
  • Trained and retrained a YOLO11 model
  • Tracking with ByteTrack and a live count panel
~/projects/drone

Computer Vision

Drone Vehicle Detection & Traffic Analytics

Detects vehicles in aerial drone footage with oriented bounding boxes and turns the detections into top-down traffic counts.

  • Rotated boxes fit angled vehicles
  • Traffic counting from a top-down view
LinkedIn demo
~/projects/basketball

Computer Vision

AI Basketball Analytics & Live Commentary

Fine-tuned YOLO to detect players, ball, rim and referees in broadcast footage, then added tracking, speed measurement and AI voice commentary.

  • ByteTrack + re-ID keep player IDs through occlusions
  • Player speed measured with homography
  • Game-event analysis with ElevenLabs voice commentary
LinkedIn demo
~/projects/ppe

Computer Vision

Employee Safety (PPE) Detection

A custom seven-class YOLOv11 model that tracks workers across site video and flags missing safety equipment.

  • Custom 7-class detection model
  • Per-worker tracking and violation flags
LinkedIn demo
~/projects/counting

Computer Vision

Traffic Analysis & Conveyor Orange Counting

One tracking pipeline for two jobs: counting vehicles on a road and counting oranges moving along a conveyor belt.

  • Unique tracking IDs prevent double counting
  • Counting-line logic for both scenes
Traffic demo Oranges demo
~/projects/fitness

Computer Vision

AI Squat & Push-up Coach

Counts exercise reps and scores form out of 100 from pose keypoints, running live on a laptop CPU.

  • Rep counting from pose keypoints
  • Form score out of 100
  • Runs in real time on CPU
Squat demo Push-up demo
~/projects/iot

Final Year Project · IoT & Backend

IoT Forest Monitoring Network

A low-cost forest monitoring network for early fire-risk awareness. Sensor nodes send temperature, humidity and CO₂ readings over 433 MHz radio to a Raspberry Pi gateway, which stores them in PostgreSQL behind a Flask REST API and dashboard.

  • Arduino nodes with DHT11 and MQ135 sensors and multi-hop relaying
  • Raspberry Pi gateway writes readings to PostgreSQL
  • Flask REST API for live data, history, alerts and users
GitHub
~/projects/action-gate
// pipeline
  1. 1Agent proposes plan
  2. 2Policy check
  3. 3Sequence risk
  4. 4Consequence replay
  5. 5ALLOW · APPROVE · BLOCK
  6. 6Audit log

Hackathon · Backend & AI Agents

ACTION GATE: AI Agent Security Gateway

A security layer between an AI agent and the business tools it can use. Every proposed plan is checked against policies and gets one decision: allow, require human approval, or block.

  • Policy enforcement and sequence risk analysis
  • Consequence replay simulates a plan in a rolled-back transaction and suggests safer alternatives
  • Audit log and Streamlit security dashboard
+1
GitHub Live demo lablab page
~/projects/insurance
// pipeline
  1. 1New claim (text, image, audio, video)
  2. 2Embeddings
  3. 3Qdrant similarity search
  4. 4Ranked similar claims
  5. 5Recommendation

Hackathon · AI & Vector Search

Multimodal Insurance Claims Processor

Team project (Stellar Engineers, 4 members). Uses Qdrant vector search across claim text, images, audio and video to find similar past claims and support fraud checks, settlement estimates and claim routing.

  • Vector search over historical claims in Qdrant
  • Similar past claims ranked to explain each recommendation
  • Streamlit interface for claim review
Live demo lablab page
More on GitHub
[05]Education

Learning & achievements.

2022 — 2026

BS Computer Science

University of Engineering and Technology (UET), Taxila

3.69/4.0

cumulative gpa

Artificial IntelligenceNatural Language ProcessingBig Data AnalyticsDatabasesArtificial Neural Networks
achievements.log
  1. 01

    Merit Scholarship

    1st position in the CS-22 batch in 4th semester, SGPA 3.95

  2. 02

    Prime Minister Laptop Scheme

    Awarded a laptop under the PM Laptop Scheme

  3. 03

    5th Position, Logic and Code Clash

    Round II, A&D Tech Innovations

  4. 04

    Aspire Leaders Program

    Cohort 4, 2026

  5. 05

    UET Women's Basketball Team

    Team member, 2023 – 2025

  6. 06

    Bronze Medal, Badminton Singles

    UET Taxila

certifications4 rows
YearCertificationIssuerLink
—Introduction to Generative AIGoogle Cloud
—Introduction to SQL & Intermediate SQLDataCamp
2026AMD Developer Hackathon: ACT III, Certificate of Completionlablab.aiView
2025AI Genesis Hackathon, Certificate of Completionlablab.aiView
[06]Contact

Let's build something useful.

Open to AI / ML Engineer roles, internships and project collaborations in machine learning, computer vision and LLM applications.

nimrafaryal2019@gmail.com
~/contact/new-message.md

Opens your email app with this message filled in.