Tom Sloboda

The Problem

Outbound marketing at scale requires aggregating contacts from disparate sources, personalising outreach, and refining campaigns from analytics — all without a team.

Architecture

• Python ingestion pipelines from multiple contact sources
• Locally hosted LLMs for personalised outreach generation
• TypeScript orchestration layer for campaign scheduling
• Analytics feedback loop for campaign refinement

Highlights

• Fully self-operated — product, engineering, infrastructure, billing
• Local LLM hosting avoids API costs at scale
• Multi-source contact deduplication pipeline

Outcome

• Operating commercially under Even Better Solutions Ltd
• End-to-end founder experience: product → infra → customers
• LLM-driven personalisation at scale with local inference

The Problem

Which ML algorithm — or ensemble — achieves the highest accuracy on facial expression classification without deep learning frameworks?

Approach

• C++ with OpenCV for feature extraction and preprocessing
• Multiple ML classifiers compared head-to-head
• Ensemble methods to boost predictive accuracy
• Rigorous experimental evaluation and reporting

Challenges

• High intra-class variation in facial expressions
• Balancing model complexity vs. overfitting
• Real-time feature extraction performance in C++

Outcome

• Ensemble methods outperformed individual classifiers
• Deep hands-on experience with ML algorithm internals
• Published on GitHub as part of MSci AI degree (Aberdeen)

The Problem

Conference scheduling is NP-hard — balancing room capacity, speaker availability, topic clustering, and attendee preferences simultaneously.

Architecture

• Java / Spring backend with scheduling engine
• Genetic algorithm for constraint-aware optimisation
• Fitness function encoding hard & soft constraints
• Web interface for schedule input and output

Challenges Solved

• Encoding multi-dimensional constraints into a fitness function
• Avoiding local optima with crossover and mutation strategies
• Performance tuning for larger conference datasets

What I Learned

• Practical application of evolutionary computation
• Modelling real-world scheduling constraints formally
• Java / Spring MVC architecture patterns