whoami

Arockiyajohn Britto

Data Scientist. Bridging the gap between data and solutions.

Birmingham, UK · MSc Data Science (Distinction)

about

cat about.txt

Data scientist with 4+ years of experience delivering machine learning, predictive analytics and AI-powered automation on large datasets. I work mostly in Python (Pandas, NumPy, scikit-learn) and SQL, and I build LLM and agentic workflows with the boring parts done properly: human-in-the-loop approvals, audit trails and model evaluation.

Before that I spent two years at NielsenIQ making sure a million-plus retail records a month were right before anyone reported on them. That habit stuck.

experience

git log --author=john

2025-09 .. now

Data Scientist & Analyst

Freelance · Birmingham, UK

  • Delivered 4-5 analysis and reporting projects for 2 clients with Python, SQL, Excel and Power BI, turning raw datasets of several thousand rows into clear insights.
  • Built machine learning and forecasting solutions, plus Python automation and AI-powered tools that cut repetitive manual work.
  • Cleaned and validated data, built Power BI dashboards for non-technical users, and ran each project from requirements to delivery.

2025-03 .. 2025-09

Advanced AI Data Trainer

Invisible Technologies · Remote, UK

  • Evaluated outputs from two frontier LLMs for accuracy, reasoning quality, relevance and instruction following.
  • Caught hallucinations, reasoning errors and edge cases, and gave structured feedback; promoted to a quality-check role on the strength of my annotation standard.
  • Wrote and refined prompts and reference responses for domain-specific tasks.

2022-07 .. 2024-08

Data Processing Analyst

NielsenIQ

  • Analysed and validated omni-channel retail and consumer data, over one million records a month, for external reporting.
  • Designed a Python and Excel VBA workflow that replaced a weekly manual process involving ~100 associates with one associate finishing in under four hours, with fewer errors and higher audit pass rates.
  • Worked on NIQ's omnichannel consumer panel built with Fetch Rewards, and on the launch of Full View™ Measurement, which extends NIQ's xAOC view across digital, physical and club channels.
  • Traced data discrepancies to root cause before downstream reporting, and reviewed peer output in internal audits.

2021-09 .. 2022-03

Data Scientist Intern

Sri Sastha Traders

  • Built demand forecasting models for regional product requirements, feeding inventory and capacity planning.
  • Consolidated sales and logistics sources into structured datasets and dashboards that flagged trends and risk areas.

projects

ls -l ~/projects

HR Onboarding Agent // Python, FastAPI, Streamlit, SQL

An AI agent that turns natural-language onboarding requests into structured actions, using LLM-based planning with confidence scoring.

  • Human-in-the-loop approvals, retry logic and deterministic fallbacks.
  • Every action lands in a SQLite audit trail.
  • Evaluated on resolution, escalation and human-override rates.

SearchSage // Python, FastAPI, LangChain, LLMs

A real-time search assistant that combines web search, result parsing and LLM answers, the way modern chat assistants reach live information.

  • FastAPI backend and Streamlit interface with prompt orchestration.
  • Answers are grounded in the parsed results, so every response stays tied to its sources.

Product Classification // scikit-learn, spaCy, Pandas

A multi-stage NLP pipeline (TF-IDF, spaCy, regex) classifying 100K+ product records across three levels of a retail hierarchy.

  • Random Forest, Logistic Regression and SVM models compared at each level.
  • 95% super group, 93% product group, 90% category accuracy, against an 85% baseline.

Data & Dev Toolkit // JavaScript, Node, Docker

Small privacy-first browser tools for data, text clean-up and development: a Python compiler, JSON formatter, regex tester, metadata viewer and more. No sign-up, no tracking, nothing leaves your device. Self-hosted, along with this site, on my own server.

publication

cat thesis.bib | head

Can AI Models Develop a Shared Sense of Ethics Through Networked Interaction?

Britto, A. (2025). MSc Data Science thesis, University of Birmingham. Supervised by Jens Christian Claussen. Zenodo.

Most alignment work tunes one model at a time. This thesis asks whether LLM agents talking to each other can arrive at shared ethics on their own. Agents sit on a scale-free network, each holding a moral vector over twelve principles; they answer moral scenarios, read their neighbours' reasoning, and change position only when persuaded. Measured with cosine similarity, entropy and PCA, convergence turns out to be possible but fragile: polarisation often persists, and biased agents in central positions pull the whole network.

  • large-language-models
  • ethical-alignment
  • opinion-dynamics
  • multi-agent-systems
  • network-simulation
  • polarisation

education

tree ~/education

2024-09 .. 2025-09

MSc Data Science, Distinction

University of Birmingham · Birmingham, UK

2019-06 .. 2022-06

BSc Mathematics, First Class

Loyola College of Arts and Science · University of Madras

tree ~/volunteering

  • Online research sales volunteer 2025 .. now
  • Student representative 2024 .. 2025
  • Health, sanitation and environment volunteer 2021 .. 2022
  • National Service Scheme volunteer 2019 .. 2022

skills

env | grep SKILLS

languages & data
Python, Pandas, NumPy, SQL, Excel, Excel VBA
ml & statistics
scikit-learn, PyTorch, TensorFlow, classification, regression, forecasting, A/B testing, feature engineering, model evaluation, NLP
generative ai
LLMs, AI agents, LangChain, LangGraph, prompt engineering, OpenAI API, function calling
cloud & deploy
AWS, GCP, Oracle Cloud, Docker, FastAPI, Streamlit, REST APIs, Git
viz & data quality
Power BI, Tableau, matplotlib, dashboards, EDA, data cleaning and validation
practices
workflow automation, data pipelines, model monitoring, data governance, auditability, human-in-the-loop systems

contact

finger john