Click to download my CV
About Me
Education
MSc Artificial Intelligence — Distinction
Queen Mary University of London
London, UK | September 2023 – August 2024
Dissertation: TinyML Fall Detection via Live Audio Classification on Microcontroller using C++, Python
BSc Computer Science — First Class Honours
Queen Mary University of London
London, UK | September 2020 – July 2023
🏆 Outstanding Academic Achievement Award 2023
Dissertation: A web e-commerce multichannel app for selling products instantaneous using Python, Django and AWS
A2 Levels
Uxbridge College
London, UK | September 2019 – August 2020
Subjects:
Work Experience
Technical Manager & Artificial Intelligence Specialist
Nationwide Utilities
London, UK | May 2024 – Present
- Designed, built and delivered the company's client-facing core platform and internal staff admin portal, end-to-end, in 5 weeks, saving £50,000+ in external costs (Next.js, React, TypeScript, Prisma, Python, FastAPI)
- Deployed AI chatbot delivering automated energy market news and insights to customers, built using AWS, Langchain, LLM APIs, and prompt optimisation
- Implemented RAG pipelines to ground LLM responses using internal market data, including context engineering, retrieval logic, and response optimisation
- Built cloud data ingestion systems on AWS, integrating multiple European gas and power market sources across Italy, Germany, France, and the Netherlands
- Led a small delivery team (developer and data analyst), providing architectural direction, code review, and delivery coordination using GitLab, Jira, and cloud deployments
- Produced HMRC-compliant R&D year report, mapping research project for tax reductions
- Managed external software partners, producing technical documentation and leading technical communication
Senior University Teaching Assistant
Queen Mary University of London
London, UK | January 2021 – June 2024
- Collaborated closely with professors to ensure seamless delivery of academic content across various modules
- Supervised and managed a team of Teaching Assistants, fostering a supportive environment
- Contributed to curriculum development and enhanced the learning experience for students
- Python Experience for 8 years (backed by various personal and industry projects)
Certificates & Hackathons
AI Engineering Professional Certificate
IBM
📅 Currently Enrolled (4th Month)
ML/DL (PyTorch, TensorFlow), GenAI/LLMs, fine-tuning, RAG agents.
1st Place Winner of QMUL AI Hackathon
Queen Mary University of London
📅 February 2025
PyTorch ML models on 10K images, 71.4% accuracy with residual connections.
Runner Up - Gemma 2 AI Challenge
Google Cloud
📅 February 2025
ShelfCare GPT: NL2SQL AI agent (2B params) for pharmaceutical inventory. React framework, fine-tuned Gemma 2 9B with PEFT/LoRA.
AWS Certified Cloud Practitioner
Amazon Web Services
📅 January 2025
AWS services: IAM, Lambda, EC2, S3, SageMaker, Bedrock, VPC architecture.
Stripe London Meetup - Developer Event
Stripe
📅 January 2025
Attended Stripe's third London meetup at their stunning Liverpool Street office. Key speakers included James Beswick on reconciling large datasets with soft deletes & tombstones to prevent data loss, Ben Smith on Stripe Workflows for building tailored experiences with complex data syncs, and Alex Casalboni on scaling SaaS pricing serverlessly with Edgee. Deep-dived into idempotency for safe retries in concurrent systems, queue-based resilience (SQS) for API throttling, and distributed system best practices.
Outstanding Academic Achievement Award
Queen Mary University of London
📅 2023
Exceptional academic performance during BSc Computer Science.
Reply Code 2021 Coding Challenge
Reply
📅 March 2021
Competed in team of 2 from QMUL to solve a 5G network optimization problem. Developed an algorithm to find optimal locations for placing antennas on a 2D map, considering signal coverage and network efficiency constraints.
Projects
FastMonitor - Enterprise SaaS Monitoring Platform
Production-grade distributed monitoring platform with microservices architecture (Node.js API, React 19 frontend, Python job checker, telemetry ingest, scheduler, alert engine). Implemented multi-provider OAuth (Google, GitHub, Apple) with TOTP-based MFA using AES-256 encryption. Built Cronitor-style hybrid push-pull monitoring with distributed locks for horizontal scaling. Real-time telemetry ingestion with sub-second latency using RabbitMQ for async multi-channel alerting (email, Slack, SMS, webhooks). Prisma ORM with 20+ tables and strategic indexing. Full Stripe payment integration. React 19 + TypeScript with TanStack Query, Redux Toolkit, i18n (8 languages). Vitest test suite with 100% coverage. Multi-language SDKs (Python, PHP). Docker orchestration with PostgreSQL, RabbitMQ, Redis.
Corrila - AI-Powered Sales Chatbot SaaS Platform
Full-stack SaaS platform for AI-powered sales chatbots with form-less lead generation. Built with Next.js 14 App Router, TypeScript, and LangChain + OpenAI GPT-4 for conversational AI. Multi-tenant architecture enabling users to manage multiple business domains with independent chatbot configurations. Real-time chat escalation using Pusher websockets for seamless bot-to-human handoff. Stripe Connect integration for subscription billing and payment processing. Clerk authentication with multi-provider OAuth (Google, GitHub, Apple). Prisma ORM with PostgreSQL managing complex relational schema (10+ models). Features custom qualifying questions, appointment booking system, email marketing, and analytics dashboard. shadcn/ui component library with 40+ components, Tailwind CSS, React Hook Form with Zod validation.
ShelfCare GPT - Runner Up in Gemma 2 AI Challenge
Built ShelfCare GPT, a Gemma2-powered AI agent that converts complex SQL queries into natural language for pharmaceutical inventory management. Achieved state-of-the-art results using only 2B parameters (runnable on mobile devices). Built React agent framework with text-to-SQL, DB overviews, and product management tools. Fine-tuned Gemma 2 9B model with PEFT/LoRA for structured outputs.
Msc TinyAED: Fall Detection Using TinyML
MSc Dissertation: Fully offline, audio-based fall detection system on microcontrollers. Achieved 93-95% test accuracy with real-time inference on Arduino and Raspberry Pi. Reduced model size by ~60% (RNN) with negligible accuracy loss.
BSc Maki: Multichannel Listing Management with Selling Agent Automation
BSc Dissertation: Full-stack multichannel e-commerce platform for individual sellers. Unified category system across platforms using recursive tree traversal and Levenshtein distance matching. Implemented OAuth flows, task-queue system for concurrent operations, real-time WebSocket updates, and automated repricing agent with neural network modelling. Deployed with containerised cloud infrastructure.
CodeRecogniser - Real-Time OCR & Automated Redemption System
Advanced multi-layered automation system built to capture and redeem Amazon gift codes displayed during live YouTube streams. Engineered custom PyQt5-based screen snipping tool with semi-transparent overlay, real-time coordinate tracking, and multi-monitor support. Integrated Tesseract OCR with custom-trained models for specialized font recognition (Minecraft-style text), achieving high accuracy on non-standard characters. Built Selenium-based web automation framework with persistent cookie-based session management for Amazon gift card redemption. Implemented sophisticated image processing pipeline converting between NumPy arrays, PIL Images, QPixmap, and OpenCV matrices. Features event-driven architecture coordinating GUI rendering, file I/O, browser automation, and OCR processing across multiple threads. Successfully automated gift card code extraction and redemption with 95% reduction in manual processing time.
TikTok Auto Uploader - 937⭐ on GitHub
Open-source automation tool for TikTok content creation and distribution. Takes YouTube URLs and automatically clips, positions, and uploads videos to TikTok, saving 30+ minutes per upload. Uses Requests library instead of Selenium for 10x faster performance. Features multi-account management, video scheduling up to 10 days ahead, CLI interface, and reverse-engineered TikTok API integration. Robust against UI changes with 937 GitHub stars and 203 forks from the developer community.
Django-Vue Auction Platform - Full-Stack ASGI Application
Full-stack auction platform with Django 3.2, Vue 3 TypeScript, and ASGI/Daphne deployment with Django Channels 4.0. Implemented dual-interval polling strategy (1s/10s) achieving 90% server load reduction with near-real-time bidding. Built 20+ REST API endpoints with <100ms response times using DRF. Automated background processing with django-crontab and multi-threaded email dispatch. Complex database with 8 models (One-to-One, One-to-Many, Many-to-Many relationships). Comprehensive security: session auth, CSRF/XSS prevention, SQL injection protection, role-based authorization, CORS, IP logging. Production-ready with Nginx, PostgreSQL, Docker, and systemd.
GeoTracker - IoT Location Tracking with AWS Serverless Architecture
Serverless IoT system tracking iPhone location events for personal analytics using AWS Lambda (Python 3.12), DynamoDB, and API Gateway. Integrated iOS Shortcuts Automation with location-based geofencing triggers to capture GPS coordinates and addresses. Implemented event-driven pipeline with Discord webhook notifications and UUID-based structured logging. Features proper IAM security, platform-specific Lambda deployment packaging, and RESTful API design. Enables commute pattern analysis and time tracking insights.
Live European Electricity Interconnector Dashboard - Serverless
Real-time electricity flow visualization dashboard for European interconnectors (France, Belgium, Netherlands, Norway) built entirely on AWS serverless infrastructure. Implemented automated data pipeline with EventBridge scheduling Lambda functions every 15 minutes, fetching from ENTSO-E API and storing in DynamoDB. Built React-Redux frontend with Chart.js hosted on S3/CloudFront CDN. Created dual Lambda architecture: scheduled data ingestion and API Gateway-triggered read operations. Configured IAM roles, SSM Parameter Store for API keys, and cross-region resource access. Optimized for AWS free tier (343,642 monthly invocations). Guided by EDF Energy company.
Energy Market Intelligence AI Agent
Production AI-driven market intelligence agent generating daily/weekly UK energy summaries. Event-driven architecture using AWS Lambda, MySQL, Docker, with LLM-powered summarization via LangChain and OpenAI APIs.
GraphRAG Legal Research Assistant - AI Knowledge Graph System
AI-powered legal document analysis implementing Microsoft's GraphRAG for solicitors. Built knowledge graph architecture with community detection, entity linking, and hierarchical organization using NetworkX. Streamlit interface with real-time chat, interactive PyVis graph visualization, and multi-format exports (TXT/DOCX/PDF). OpenAI GPT-4 integration for semantic search and embeddings. Performs multi-hop reasoning across legal document corpus beyond traditional vector similarity RAG. Docker containerized.
COOL Compiler - Complete 4-Phase Implementation from Scratch
Full-featured compiler for COOL (Classroom Object-Oriented Language) implementing all classical compilation phases in Java. Designed custom ANTLR 4 lexer/parser grammars for OOP constructs, inheritance, and pattern matching. Built semantic analyzer with symbol table management, type checking, and scope resolution using Visitor pattern for AST traversal. Implemented MIPS assembly code generator with stack machine, object layout, and method dispatch tables. Professional build system with modular compilation and comprehensive test suite comparing against reference compiler. Demonstrates graduate-level compiler theory.
Bare-Metal Embedded Firmware for ARM Cortex-M0+
Developed bare-metal embedded firmware in C for NXP FRDM-KL25Z development board (ARM Cortex-M0+ microcontroller). Implemented direct register-level I²C peripheral configuration (400 kHz fast mode) to interface with character LCD module. Configured hardware pin multiplexing (PTC1/PTC2), clock gating, and peripheral initialization from scratch without abstraction layers. Implemented LCD initialization sequence following datasheet specifications with timing-critical command transmission. Built using Keil µVision 5 with CMSIS support, demonstrating professional embedded boot structure: system clock setup, UART debugging, I²C bus configuration, and device initialization. Successfully validated I²C communication and LCD control through register-level programming.
Cognitive Robotics: Human Action Prediction from First-Person Vision
Deep learning system predicting hand movement direction (left/right) from egocentric video using custom AlexNet architecture trained in PyTorch. Implemented full ML pipeline: OpenCV video preprocessing with spatial cropping, keyframe extraction (50 frames/video), custom PyTorch Dataset with data augmentation, and CNN training with SGD optimizer and learning rate scheduling. Built forward hooks for layer visualization, generating Grad-CAM-style attention heatmaps to interpret model decisions. Addressed small dataset challenge (~900 images) through dropout regularization, weight decay, and early stopping. Applied cognitive robotics research concepts for anticipatory action prediction in human-robot collaboration scenarios. Demonstrates end-to-end ML engineering from raw video to interpretable model predictions.
SOCKS5 Proxy Server - Published on PyPI
Custom-built SOCKS5 proxy server implementation using Python sockets, eliminating the need for VPN subscriptions. Published as PyPI package (makiproxy5) for easy installation. Built by reverse-engineering the SOCKS5 protocol through RFC documentation and Wireshark packet analysis. Features authentication support, concurrent client handling, secure mode, and configurable host/port settings. Originally deployed on Raspberry Pi to access UK TV content internationally. Demonstrates deep understanding of network protocols, socket programming, and practical reverse engineering.
TFL Route Planner
Given start and end destination, and using DFS, UCS, BFS and A* with custom heuristic definition, will find optimal path to destination station. The algorithm considers time for each station, as well as transition time between different tube lines.
Cryptographic Algorithm Implementations - 3DES & LFSR
Pure Python implementations of fundamental cryptographic algorithms from scratch. **Triple DES (3DES)**: Industry-standard symmetric-key block cipher with 192-bit key size (64-bit × 3), implementing DES permutation tables, S-boxes, and key scheduling following NIST standards. **Linear Feedback Shift Register (LFSR)**: Stream cipher implementation for pseudorandom bit generation, featuring 8-bit seed generating up to 255 bits for encryption/decryption. Demonstrates practical understanding of both block ciphers and stream ciphers, XOR operations, feedback polynomials, and cryptographic period analysis. Built inspired by Christof Paar's cryptography lectures, showcasing deep knowledge of symmetric encryption primitives and security algorithm design.
Internet Networks and Protocols Video Series
23-hour comprehensive video series teaching content from the 800-page "Computer Networks: A Top-Down Approach" textbook. Detailed explanations of protocols, infrastructure, and network concepts covering all 5 OSI layers, Cellular Networks (4G LTE, NGE), and Telephone Networks.
Stock Market Investor Simulator
Java-based stock market trading simulator game demonstrating advanced object-oriented programming principles. Implemented polymorphism for different stock classes, abstraction through abstract classes, inheritance hierarchies, and encapsulation for data security. Features include real-time portfolio management, random market events affecting stock prices, multi-currency support with conversion fees, persistent storage using custom .maki file format, and comprehensive buy/sell functionality across multiple market sectors (Energy, Healthcare, Consumer, Industrial).
StockerAnalysis - Financial Statement Analysis Tool
Python financial analysis tool processing 4 years of company data via yfinance API. Calculates profit margins, asset-liability ratios, and cash flow metrics with matplotlib visualizations. Provides free insider trading data for US stocks (typically premium content). Parses Yahoo Finance portfolio CSVs for sector distribution and market cap breakdown. Implements Benjamin Graham's "Intelligent Investor" defensive criteria. MongoDB integration for persistence. Democratizes Bloomberg-level analysis.
Big Data Wash Trading Detection - Ethereum Blockchain Analytics
Processed 370M+ Ethereum transactions (Aug 2015 - Jan 2019) using Apache Spark and PySpark to detect fraudulent wash trading patterns. Engineered MapReduce pipelines with optimized reduceByKey and join operations, reducing memory overhead for petabyte-scale data. Implemented graph-based fraud detection using Directed Acyclic Graphs (DAGs) to identify circular transaction patterns. Built cloud infrastructure with AWS S3 and Boto3 for distributed storage. Developed time-series analytics identifying top smart contracts by Ether received and top miners by block size. Created Jupyter Notebook visualizations analyzing blockchain storage overhead.
Property Listing Webapp - Zoopla Clone
Full-stack property listing platform with advanced filtering capabilities. Features include MongoDB-hosted property database, detailed property views with multiple images, admin and user dashboards for account management, landlord contact functionality, and comprehensive property information (price, bedrooms, bathrooms). Built as a university group project with leadership role.
