Stories

---

While studying for my MSc in Biotechnology at the University of Greenwich, I started thinking about which area of biotechnology I wanted to specialise in. During that time, I became fascinated by microalgae research. These tiny organisms can grow using carbon dioxide, sunlight and a small number of nutrients, and they have played a fundamental role in Earth's history by producing oxygen since the earliest stages of life. I was equally captivated by their potential as sustainable cell factories capable of producing valuable compounds such as lipids and natural pigments.

This interest led me to begin a funded PhD investigating the production of colourless carotenoids from a specific microalgal species. These compounds act as natural UV filters and have potential applications in sunscreens, cosmetics and nutritional supplements. I was particularly drawn to this research because it combined biotechnology, molecular biology and analytical chemistry to explore compounds with real-world applications.

My research also explored β-carotene, a related carotenoid with potential applications in preventing or slowing the progression of retinal diseases such as Retinitis Pigmentosa. Having seen my grandmother affected by this condition, I felt a personal connection to this part of the research and found it especially meaningful.

My PhD was primarily laboratory-based. I spent much of my time designing experiments, cultivating microalgae and characterising pigments using a range of analytical techniques. Research often involved careful planning, analysing results, troubleshooting experiments and repeating procedures to improve reproducibility. Although I enjoyed the laboratory work, I found myself becoming increasingly interested in analysing and interpreting the data generated from the experiments.

That growing curiosity eventually led me to explore scientific software and data analysis tools, before ultimately deciding to learn Python.

Looking back, this marked the beginning of my transition from experimental biotechnology towards computational biology, bioinformatics and, later, artificial intelligence. Although my interests have expanded beyond laboratory research, I remain deeply interested in microalgae biotechnology and still see myself contributing to this field in the future. One of my goals is to combine my scientific background with computational approaches to help tackle meaningful challenges in biotechnology.

---

Back in 2022, after completing the laboratory work for my PhD, I had some free time while working part-time and writing up my thesis. I decided to embark on a learning journey in Python through Angela Yu's Udemy course.

I quickly realised that programming had many similarities with scientific research: both require breaking down complex problems or objectives into smaller and smaller steps until they become manageable.

I started with the fundamentals, learning concepts such as functions, loops, conditional statements, lists and dictionaries, before moving on to more intermediate and advanced topics, including object-oriented programming, list and dictionary comprehensions, JSON parsing and working with APIs.

The course also introduced me to a wide range of applications, from data analysis and visualisation with pandas and other Python libraries to web development with HTML, CSS and Flask.

By the end of the course, my previous perception of programming had completely changed. What had once seemed intimidating became much more approachable, and I had built a solid foundation that allowed me to start developing small projects independently, including this website.

---

It was quite an adventure to work as a contractor for Google DeepMind, collaborating with AI engineers as a Biology Expert during both the training and post-training stages of frontier language models.

My role ranged from performing loss analyses to understand where the models struggled within specific areas of biology, to designing educational content such as simulations, animations and interactive scenarios that could be used during training. In practice, much of the work involved prompt engineering—or what is increasingly referred to today as system engineering—to improve model performance and reasoning.

Working in such a fast-paced environment taught me the importance of delivering high-quality work under tight deadlines while collaborating with multidisciplinary teams across different time zones, particularly in the United States and India. It also gave me a unique insight into how frontier AI models are developed and evaluated.

Perhaps most importantly, the experience sparked a deeper curiosity about large language models and agentic systems. It reinforced my desire to continue building at the intersection of biotechnology and AI, a direction that has since shaped many of the projects featured on this website.

---

I have always been fascinated by bioinformatics, and the HackBio AI for Genomics programme gave me the opportunity to work with a real biological dataset containing drug response measurements across cancer cell lines. The project combined data analysis, biological interpretation and machine learning, allowing me to explore how computational methods can be used to better understand complex biological data.

During the first stage, I worked collaboratively as part of a team to explore, clean and analyse the dataset using Python. In the second stage, I independently developed and evaluated machine learning models to predict drug response, gaining practical experience in feature engineering, model evaluation and biological data interpretation.

One aspect I particularly enjoyed was seeing how programming, statistics and biology come together to answer biological questions. The project strengthened my interest in bioinformatics and gave me practical experience of the complete workflow, from exploring raw biological data to building predictive machine learning models.

---

I wanted to reproduce an RNA-Seq analysis pipeline to learn each step of the process, from Bash scripting and project organisation to gene expression analysis and data visualisation.

To make the project more meaningful, I combined it with my background in microalgae and chose a topic that is familiar to me: the effects of nitrogen starvation on cellular metabolism and gene expression. I selected Chlamydomonas reinhardtii as the model organism because its genome is well characterised and extensively documented.

Throughout the project, I used frontier AI models as learning assistants to guide me through each step, together with the knowledge I had previously gained from the HackBio bioinformatics courses. Rather than simply accepting the outputs, I made a conscious effort to understand each stage of the pipeline, ensuring that the analyses made biological sense and supporting my understanding through literature searches.

The aim of this project was not simply to reproduce an existing workflow, but to understand the rationale behind every step and build a solid foundation in RNA-Seq analysis.

---

After learning how to build AI agent workflows with n8n and other low-code platforms through Ed Donner's AI Builder course, it felt natural to apply those concepts to a domain I knew well from my PhD: microalgae biotechnology.

I decided to build a Retrieval-Augmented Generation (RAG) knowledge assistant for a fictional microalgae company. To simulate a realistic research environment, I designed a collection of synthetic datasets representing microalgae strains, engineered strains, cultivation experiments and commercial products, organised across multiple spreadsheets.

Using these datasets, I created a RAG knowledge base by generating vector embeddings and storing them in a Supabase vector database. I then developed an AI agent in n8n and connected it to a simple Flask web interface, allowing users to interact with the knowledge base through natural-language questions.

Developing the workflow involved a significant amount of R&D to optimise the retrieval pipeline, improve response quality and understand how different design choices affected the assistant's performance. I validated the system using questions of varying complexity, including paraphrased queries and deliberately incorrect questions to evaluate how reliably it retrieved and interpreted the underlying scientific information.

It was particularly rewarding to combine my background in biotechnology with modern AI tools and build something that could potentially support researchers working with scientific data.

---

During the 2025/2026 season, I had the privilege of working in one of the VIP lounges at London Stadium, the Great Briton Lounge, during West Ham home matches. It was also quite a unique experience to be around former football legends, current players and the atmosphere that comes with Premier League matchdays.

One thing that struck me was that, on several occasions, fans and customers seemed confused about which ticket package they had purchased, particularly whether their package included the buffet or not. From what I observed, this sometimes slowed down the flow of customers entering the lounge and added pressure to the hospitality team. The ticket itself did not clearly show which hospitality package it came with, presumably because the stadium ticket and the hospitality package were managed through different companies.

I felt that something could potentially be done to make this experience clearer and improve the customer journey. I kept these observations at the back of my mind for several months.

In early April 2026, I started learning AI Engineering through Ed Donner’s course on Udemy. By the way, it is an amazing course for anyone interested in learning about AI Engineering: very clear, practical and engaging. Through the course, I learned about large language models, tool calling, structured data and how AI assistants can interact with external systems.

This gave me the idea of building an AI-powered assistant that could help customers explore and compare the available ticket packages. The assistant clearly presents the differences between basic, lounge and all-inclusive options, allowing users to understand what each package includes before making a selection.

However, this project represents only one possible and relatively simple solution. It should be seen as an initial prototype or first draft rather than a fully developed answer to the problem.

More upstream research would be needed before refining the solution. This would include understanding the full customer journey, identifying which organisations manage the different parts of the ticketing and hospitality experience, speaking with the relevant stakeholders and gathering more direct feedback from customers.