Tirth

AI/ML Engineer

|

About

Profile picture

Here is a little background

Hey there! I’m a Software & Machine Learning Engineer based in the UK, with roots in India and a strong foundation in Computer Science. I hold a Master’s degree in CS with a focus on Human-Computer Interaction from the University of Birmingham. I’m passionate about using AI and machine learning not just to solve tough problems, but to build thoughtful, user-centered solutions that make a real impact. Whether it’s writing clean, scalable code or collaborating on open-source projects, I love blending technical depth with human-focused design. Currently, I’m exploring new opportunities where I can combine my skills in AI, ML, and software engineering to build innovative, meaningful tech. Let’s connect if you're working on something exciting—or just want to geek out over cool ideas!

Experience

Research Associate

University of Birmingham

Oct 2024 - Jan 2025

Python
OpenCV
NetworkX
spaCy
HuggingFace
TensorFlow
  • Developed a pipeline to process Accimap diagrams (PNG/PDF), converting images into structured data for easier incident analysis.
  • Applied AI techniques to study connections in Accimaps, uncovering how different system levels interact.
  • Enhanced analysis by processing text, building graphs, and visualizing interactions with clustering techniques.

Full Stack Intern

Our Time HQ

Jun 2024 - Aug 2024

HTML
CSS
JavaScript
React
Python
FastAPI
PyTorch
spaCy
pandas
Git
Docker
NumPy
  • Built an employee time management calculator, designing an easy-to-use interface and a solid backend to track work hours and measure productivity.
  • Assisted in creating Nexie, an AI-powered time management assistant, by researching how users manage their time to help the AI better understand their needs.
  • Collaborated with different teams to connect the calculator with Nexie’s AI features, delivering tailored time management insights for users.

AI Engineer

Inzeitech Digital Solutions

Oct 2023 - Apr 2025

Streamlit
Figma
Python
HTML
Java
JavaScript
React
  • Delivered comprehensive AI solutions for clients by building LLM-powered applications with retrieval systems and semantic search, addressing specific business challenges
  • Developed practical AI applications utilising vector databases and optimised embeddings, creating testing frameworks that enhanced prompt quality and system performance.
  • Spearheaded innovation by implementing advanced AI methodologies, significantly reducing response times and enhancing user experience across various client projects.

Skills

Hover over a skill for current proficiency

Streamlit

80%

HuggingFace

90%

CSS

80%

NetworkX

75%

FastAPI

80%

Groq API

85%

NumPy

85%

TypeScript

80%

Java

90%

LangChain

85%

PostgreSQL

70%

spaCy

75%

pandas

85%

Ollama

90%

LangGraph

85%

PyTorch

90%

JavaScript

85%

Docker

80%

Pillow

75%

Scikit-learn

80%

Figma

95%

Git

90%

OpenCV

80%

TensorFlow

80%

Python

90%

HTML

80%

Qdrant

80%

React

85%

Projects

GraphMinds - MSc Dissertation Project

Project 1: GraphMinds - MSc Dissertation Project

Python
NetworkX
Scikit-learn
pandas
Ollama

Developed GraphMinds, a security-focused AI system that combines Large Language Models with Knowledge Graphs to analyse complex data locally. Designed an innovative approach for processing unstructured information by mapping both direct and indirect relationships between entities in knowledge graphs. Implemented the entire system locally using Sentence Transformers for embeddings, NetworkX for graph visualisation, and Ollama for LLM integration, ensuring data confidentiality without cloud dependencies. Project successfully enhances analysis of fragmented datasets, particularly valuable for knowledge-intensive applications requiring comprehensive relationship mapping. Completed under supervision of Prof. Christopher Baber as part of my MSc in Human-Computer Interaction at the University of Birmingham.

RAG using NLP

Project 2: RAG using NLP

Python
spaCy
JavaScript
Docker
Git

Designed and implemented a high-performance RAG pipeline that enables accurate question-answering from PDF documents while maintaining complete data privacy through local GPU acceleration. The system processes documents up to 500 pages using sophisticated NLP techniques—including semantic chunking and vector-based retrieval with FAISS—achieving 87% information retrieval accuracy with reduced hallucinations. Developed with a modular architecture supporting multiple LLM providers (OpenAI, Gemini, Groq, and local models), the project features an intuitive FastAPI web interface with Server-Sent Events for real-time processing feedback, resulting in a 40% reduction in query-to-answer latency through advanced parallel processing techniques. The implementation demonstrates expertise in machine learning, NLP, GPU optimization, and modern software engineering practices while delivering a production-ready solution for private document intelligence.

View on GitHub

Brain - Computer Interface

Project 3: Brain - Computer Interface

Python
PyTorch
Streamlit
Docker

Developed a brain-computer interface to decode speech sounds from EEG signals, training a model to recognise eight distinct speech phonemes with a 92.5% F1-score on synthetic data, while creating an interactive app that displayed topographic scalp maps and used visualisation techniques to show how the model interprets brain activity, making it easier for researchers to analyse results in real-time and support cognitive studies, with added reliability through cross-validation, hyperparameter tuning, and data augmentation.

View on GitHub

ResearchWebGraph

Project 4: ResearchWebGraph

FastAPI
Streamlit
Groq API
Qdrant
spaCy
LangChain
LangGraph
Python

Built an AI-powered tool to help researchers tackle academic papers more efficiently, letting them upload PDFs to run semantic searches, extract key entities like concepts, authors, and references, and even answer specific questions about the content using a retrieval-augmented generation pipeline, while also creating knowledge graphs to map out relationships between studies, summarize papers, and highlight connections, all wrapped in a straightforward interface that made these insights easy for nontechnical users to dive into, boosting engagement by 45% among 10 beta testers.

View on GitHub

MoodMoji

Project 5: MoodMoji

Python
TensorFlow
OpenCV
Pillow

Developed a real-time facial emotion recognition app that analyzes webcam video to detect emotions and pairs them with custom emojis, training a convolutional neural network on the FER2013 dataset to classify seven emotions—angry, disgust, fear, happy, neutral, sad, and surprise—and designing an interactive GUI to show the live video feed alongside matching emojis, highlighting skills in deep learning, computer vision, and interface development.

View on GitHub

Contact

Need a creative mind? Look no further! Let's join forces and make magic happen. Coffee's on me – unless you prefer tea!

+44 7741918549

tirthkanani18@gmail.com

London, United Kingdom

Crafted with ❤️ by Tirth