From raw data to deployed models — I design, train, and integrate machine learning systems. Specialised in deep learning, LLM-based agents, and reinforcement learning.
CNN-based classification, object detection, and segmentation. Trained models for plant disease detection, phishing site analysis, and aerial imaging.
Fine-tuning language models, building RAG pipelines with LangChain and LlamaIndex, and chaining LLM calls for structured reasoning.
Multi-step agent workflows using LangGraph. Agents with tool use, memory, and planning capabilities for complex real-world tasks.
Q-learning agents for sequential decision-making. Applied to autonomous drone navigation and game-playing agents.
End-to-end pipelines from raw ingestion through preprocessing, feature engineering, model training, and serving.
Wrapping trained models in FastAPI or Django endpoints, with input validation, versioning, and response formatting.
Understanding the problem domain, exploring the dataset, and deciding between supervised, unsupervised, or RL approaches.
Running baseline experiments, comparing architectures, tuning hyperparameters, and tracking metrics with proper tooling.
Training on GPU, evaluating on held-out test sets, and diagnosing overfitting, underfitting, and class imbalance.
Exporting models, building inference APIs with FastAPI, and integrating into web applications or pipelines.
CNN-based leaf disease classifier trained on 38 plant classes with high accuracy.
Q-learning agent trained to navigate a drone and trigger aerial photography at optimal waypoints.