Devis Nishku

Specialisation in AI/ML. Core coursework: Neural Networks and Deep Learning, Natural Language Processing and Streaming Data Science.

Core coursework: Algorithms and Data Structures, Software Engineering, Web Information Technologies and Databases.

Observe layer of an AI workflow-transformation platform: LLM-based ingestion of interview transcripts, emails, policy documents and system logs, reconciled into a canonical as-is workflow graph with typed edges and per-node provenance. Detects "say-vs-do" mismatches between stated and actual processes via evidence triangulation.

Autonomous agent playing a "Who Wants to Be a Millionaire"-style quiz under strict constraints: 30s per answer, Colab T4 memory budget, no paid APIs. Built on open-source Qwen2.5-7B-Instruct with quantization and prompt engineering; designed the web-search-based RAG pipeline for highly specific, non-general-knowledge questions.

Native integration of Continuous LSTM (cLSTM) and continuous Progressive Neural Networks (cPNN) as classifiers inside the CapyMOA streaming framework. Anytime inference with sliding windows and dynamic column expansion for concept drift.

End-to-end data cleaning, quality assessment and enrichment pipeline on a real-world messy dataset. Fuzzy record linkage, schema validation and quality scoring across 6,000+ records.

Breast cancer tissue classification using fine-tuned EfficientNet, ResNet, and Vision Transformer (ViT). Applied Grad-CAM for model interpretability.

Multivariate time-series classification using LSTM, GRU, and hybrid architectures. Explored attention mechanisms and temporal convolutions for sequence modelling.

Built applicant tracking system backend with Java Spring Boot, Apache Lucene, and PostgreSQL. Focused on full-text search and CV processing workflows.