Resume

Perusha Moodley

Contact Information:

• Location: Reading, UK

• LinkedIn: linkedin.com/in/perusha-moodley

• Website: www.perusha.dev

Professional Summary

Highly experienced developer and technical team lead with over 18 years in software development and consulting. Recent PhD in Deep Reinforcement Learning, now pivoting to AI Safety and alignment. Currently enrolled in Blue Dot’s AI Safety Fundamentals Alignment course and an AISC (AI Safety Camp) project (Jan-April 2025). Lifelong learner adept at analysing and coding research papers and adopting new technologies. Proven track record in leading projects and developing innovative solutions. Seeking impactful role working on AI safety and alignment problems including technical governance for the model development-deployment evaluation process.

Core Competencies

• PhD in Machine Learning, specialised in Deep Reinforcement Learning (RL/DRL)

• Skilled in Python, PyTorch, LLMs

• AI Safety & Alignment

  • Blue Dot Alignment course - October 2024 intake

  • Blue Dot project on CAI - Constitutional AI. 

  • AISC project team working on an AI Safety Scientist 

  • Interpretability of transformers for RL tasks

• Agent based development (LlamaIndex and some LangChain)

• Software Development & Business Analyst

• Leadership & Team Management

• Consulting & Client Engagement

Professional Experience

Lodgical Ltd - AI Research Development

Role: Founder & Lead Consultant
Duration: 2006 - Present

• Managing a consultancy specialising in MLOps, AI research, and software development.

• Consulted on projects and architected integration pipelines and developed applications for clients across sectors, with significant experience in the QA critical pharma sector.

• Designed and developed algorithms using PyTorch (GPU); contributed novel insights into transfer and generalisation using unsupervised methods and to tokenization and position encoding for multimodal tasks in transformers, with some interpretation analysis.

• Currently developing skills in AI safety and alignment, aiming to provide assistance for model evaluations and technical governance process flows.

Baxter Healthcare Ltd — Senior Integration Specialist

Duration: 2006 - 2017

• Designed and developed integration frameworks, including logging and monitoring, in enterprise environments for automated warehousing/CRM systems. Deployment of code from unit, functional and regression testing to highly regulated production systems in the pharma industry.  

• Technical team lead for EU-wide projects

• Business analyst for supply chain, warehousing, and financial processes.

• Developed backend applications using C++/Java (Weblogic servers), SQL databases (Oracle), and enterprise messaging tools (Message Queues).

Forza Consulting BV - Senior Technical Consultant

Duration: 2010 - 2015

• Co-designed a product for automating payments using OCR integrated with ERP systems (Web services/Weblogic server).

UTC FS - Development Lead

Duration: 2005 - 2006

• Development Lead responsible for global development standards and support.

Baxter Healthcare Ltd - Senior Analyst Programmer

Duration: 2001 - 2005

• Developer and functional analyst for finance, supply chain procurement and planning during EU rollout.

Deloitte and Touche - Senior Consultant

Duration: 2000 - 2001

• Provided consulting on business systems.

PwC - Consultant

Duration: 1998 - 2000

• Provided consulting on business systems, leveraging expertise as a developer and whitehat hacker.

Education

Ph.D. in Computer Science
University of Reading, UK
Focus: Deep Reinforcement Learning (RL/DRL)

M.Sc. in Mechanical Engineering
University of Natal, South Africa
Focus: Manufacturing automation

B.Sc. in Mechanical Engineering
University of Natal, South Africa
Focus: Robotics and automation

Research & Publications

• Interpreting Decision Transformer: Insights from Continuous Control Tasks, Accepted for Conference 2024.

• Multi-State-Action Tokenisation in Decision Transformers for Multi-Discrete Action Spaces, arXiv [Cs.LG], Pending Review.

• A Conservative Q-Learning Approach for Handling Distribution Shift in Sepsis Treatment Strategies, NeurIPS 2021 Workshop.

• Understanding structure of concurrent actions, Springer 2019

Current Projects

• Enrolled in the AI Safety Fundamentals AI Alignment course from BlueDot (Oct 2024 - Feb 2025).

• Blue Dot project is on Constitutional AI - CAI

• AISC 2025 Project team: AI Safety Scientist (Jan 2025-April 2025)

Additional Activities

• Member of the steering committee for the Thames Valley AI Hub (TVAI), an initiative supporting collaboration between academia, industry and enthusiasts.  

• Teaching Assistant for ML post-graduate courses.

• Moderator for ML Collective RL group; organised paper reading groups and coding sessions.

• Former organiser of Google Developer Group events from 2015-2021; grew group to >2.5k people, ran monthly technical talks and workshops; ran yearly conferences attracting up to 300 people. Actively helped people skill-up and re-train for new careers.

• Mentored women in tech.

• Early contributor to the OpenMined open-source project.

Technical Skills - Detailed

I have designed, modified and debugged algorithms both in RL/DRL and more generally in ML using PyTorch (GPU). I have extensive experience with both decision transformers and core RL algorithms. In the former I made novel contributions to the core transformer algorithm’s tokenisation and position encoding for multimodal RL tasks with visual inputs, with a view towards improving interpretability. Part of my work involved optimising the dataloader for the large multimodal datasets we trained on and tuning models. I wrote methods from mechanistic interpretability, used in AI alignment, to retrieve model activations and analyse model behaviour. In the latter, I worked with both on-policy (PPO) and off-policy (DQN) algorithms, in multi-task settings and with auxiliary signals. 

While my primary experience is in RL, I have experience working with clustering, contrastive methods, ResNets and LLMs (fine-tuning and LlamaIndex). I understand the transformer architecture very well and have trained models from scratch using GPUs with PyTorch. I have used multiple frameworks including SB3, CleanRL, Ray and RLLib and Gym/Gymnasium (created custom environments). I worked on projects extending the interpretation of decision transformers and skill transfer for transformer RL agents. 

Prior to my PhD I worked as an enterprise developer and analyst so I am experienced with working with users and on production systems. I held development team lead positions with responsibility for setting standards, best practices and architecture design.

Finally and most importantly, I am a life-long learner. As I have demonstrated throughout my career, I remain strongly interested in learning and confident in my ability to pick up any new skills.