Skip to content

How this started

irit began in November 2026 as a side project between a software engineer, Dzulfikar Ats Tsauri (industrial software, Rust, IoT), who builds the tools, and a microbiologist, Naufal Rafif Ramadhan, heading into graduate work on environmental microbial ecology and genomics, who validates them. Neither wanted to start a biotech company; the aim was to build something useful and keep current with computational biology.

Six directions came up and were compared on research demand, novelty, feasibility and how hard the results are to validate:

Idea Verdict
AI workflow agent for bioinformatics pipelines Easy to build, crowded (Biomni, BioMaster, Seqera AI, Galaxy MCP). Kept as a later layer on top of the tools, not the product.
Sequence analysis workbench Mature free alternatives. Dropped.
Molecular docking / MD workbench Compute-heavy and hard to validate. Dropped.
Multimodal microbial observatory (sensors plus periodic sequencing) Strong fit for IoT skills, but needs field partners and years of data. Long-term.
Microbial interaction simulator (“bacteria battles”) Kept as a small outreach demo built on published two-species models, not a research product.
AI virtual cell (inspired by GenBio’s AIDO Cell) Multi-year research needing perturbation data that doesn’t exist for most microbes. Out of scope.
  1. Research over opinion. Rather than build what one person found annoying, we surveyed issue trackers, forums and papers for where researchers lose time (landscape). The strongest signals were sample sheets, reference databases and QC aggregation: manual steps, not missing science.
  2. The laptop constraint. Naufal works on a modest laptop, and so do many students. Running a standard pipeline locally showed the real bottlenecks were disk (16 GB of containers) and memory-hungry tools (FastQC, GTDB-Tk), not CPU. That became the organising principle: tools that work on the computer people actually have.
  3. Measured claims only. Each tool is checked against the established tool it replaces or feeds before any number is published here.

GenBio AI’s AIDO Cell (previewed August 2026) simulates two human cell lines (K562, HepG2) as a persistent state you can perturb step by step. Its benchmark is self-reported and access is limited to collaborators. The idea of a “virtual microbe” is appealing for microbial ecology, but such models learn from large perturbation experiments, which can’t be run on organisms that can’t be cultured, including much of environmental microbial diversity. We treat it as a direction to watch, not build.