The inability of DNA sequencing technologies to interpret entire molecules led to the development of methods that connect the obtained short fragments back together in a puzzle-like process. They are called assemblers and their design is guided with the notion that similar fragments originate from the same genomic region. That is often annulled due to sequencing errors and repetitive nature of the genome. The advent of new sequencing approaches, namely Pacific Biosciences and Oxford Nanopore Technologies, pushed the limit on the fragment lengths at a cost of higher error rates, but still facilitated the assembly problem considerably. First assembly attempts used various types of error correction approaches prior the assembly with existing tools at that time. Although, several long read based assemblers have been proposed in the past years, they demand significant amounts of computational resources. The focus of this research is development of memory efficient and scalable algorithms for de novo assembly of large genomes using raw third generation of sequencing data. In the scope of the thesis we implemented three novel tools for genome assembly: a memory friendly layout module called Rala, which builds the assembly graph from preprocessed sequences and resolves junctions in graph with the help of force directed placement; a fast and accurate consensus module called Racon based on vectorized partial order alignment; and the complete de novo assembler called Raven, which competes with state-of-the-art assemblers both in quality and resource management.