Skip to content
PrimeOmicX

00 · Computational Biology × Applied AI

From biological data to intelligent systems.

PrimeOmicX is an independent computational biology and applied-AI practice founded by after nine years of omics research and pipeline engineering in academic and public-health settings. It builds reproducible workflows, agentic AI systems and small language models that run on the hardware you already own.

Start a projectRead the services

  1. Raw data

    FASTQ, BAM, mass-spec, arrays

  2. Pipelines

    Reproducible processing

  3. Analysis

    Statistics & interpretation

  4. Knowledge

    Indexed, searchable, cited

  5. Agents

    Work that runs itself

  6. On-device models

    Private, offline, fast

Figure 1 | Pipelines make analysis repeatable. Analysis becomes knowledge you can search. Knowledge powers agents. Agents run on small models. Hire any one stage, or the whole path.
Research record
The founder's 18 peer-reviewed papers, including Cell, Nature Immunology and PNAS
Public work
The founder contributed to aquascope, an open-source wastewater surveillance pipeline
On your hardware
A 3B model in 2 GB, offline on any Apple Silicon Mac

Materials. Nextflow, nf-core, Snakemake, GATK, DeepVariant, STAR, Salmon, DESeq2, Seurat, Scanpy, Cell Ranger, Kraken2, MOFA+, Docker, Apptainer, SLURM, AWS Batch, Seqera, PostgreSQL, pgvector, Qdrant, LangGraph, MCP, PyTorch, MLX, llama.cpp, Core ML, Hugging Face.

01 / Services

Five practices, one stack.

Each one stands alone. Together they take a project from the sequencer to a deployed model.

  1. 01

    Omics Analysis

    Analysis you can defend in a review.

    From raw data to publication-ready figures, with the methods, code and environment needed to reproduce them.

    WGS / WES · Bulk & single-cell RNA-seq · ATAC / ChIP / methylation · Metagenomics · Multi-omics

  2. 02

    Pipeline Engineering

    Pipelines that still run in five years.

    Production-grade Nextflow, Snakemake and WDL workflows that run the same on a laptop, a SLURM cluster and the cloud.

    Nextflow / nf-core · Snakemake · WDL · Containers · HPC & cloud

  3. 03

    Agentic RAG Systems

    Answers with sources, not vibes.

    Question answering over your papers, protocols, LIMS and results, grounded in citations and deployable on-prem.

    Hybrid retrieval · Live database tools · Citations · Evaluation · On-prem

  4. 04

    AI Agents

    Agents that do the work, and show their work.

    Tool-using agents that run analyses, triage QC failures, draft reports and file the paperwork, with humans in the loop.

    Pipeline operators · QC triage · Report drafting · MCP tools · Guardrails

  5. 05

    On-device AI Models

    Frontier-quality answers, laptop-sized models.

    Small language models, around 3B parameters, fine-tuned on your domain and running on Apple Silicon. No data leaves the machine.

    ~3B parameters · Fine-tuning · MLX / llama.cpp / Core ML · Offline · macOS apps

02 / On-device models

Small models, benchmarked against the large ones.

Below is a scripted run of a fine-tuned 3B model answering over a local index. Nothing leaves the laptop, and there is no per-token cost.

Size
About 3B parameters; 4-bit weights fit in roughly 2 GB
Trained on
Your protocols, reports and terminology
Runs
Fully offline on any Apple Silicon Mac, no per-token cost

Box 1 | Question answering on a MacBook, no network

run: primeomicx-3b · 4-bit · MLX · MacBook Pro, M4

>

tokens
0
tok/s
—
peak memory
1.9 GB
bytes leaving device
0
Box 1 | Illustrative transcript. The output is scripted; the timings are representative of the configuration named above, not a benchmark.

More on on-device models →

03 / Pipelines

A pipeline is software.

Built the way you would want to inherit it: one repository, tests, containers and CI, running the same on a laptop, a SLURM cluster or cloud batch. Figure 2 is a public pipeline the founder worked on before PrimeOmicX, shown as an example of that standard.

Handover
Runbooks and a walkthrough for the next maintainer
Tested
nf-test suites and CI on every change
Benchmarked
Runtime and cost measured on your data
aquascope · 14 processes · 3 stagesNextflow DSL2 · CDCgov/aquascope
  1. 1

    Pre-processing

    Ion Torrent, Illumina, Nanopore and PacBio are split by read length, checked with FastQC or NanoPlot, and trimmed with fastp.

  2. 2

    Alignment

    Trimmed reads are aligned with Minimap2; already-aligned Ion Torrent BAMs are re-headered instead.

  3. 3

    Trimming and variant calling

    Both paths go through iVar primer trimming and sorting, then Freyja assigns lineages and iVar estimates variants. Every step reports to MultiQC.

Figure 2 | aquascope, an open-source SARS-CoV-2 wastewater surveillance pipeline developed at CDC for the National Wastewater Surveillance System, redrawn from the project's documentation. The founder contributed to it before starting PrimeOmicX. It is CDC's work, shown here as background: it is not a PrimeOmicX product or engagement, and its inclusion implies no endorsement. Published in Emerging Infectious Diseases, 2025; source at github.com/CDCgov/aquascope.

More on pipeline engineering →

04 / Methods

Five steps, agreed in writing.

  1. 01

    Discovery

    A scoping call and a look at your data, infrastructure and constraints. You get a one-page plan with milestones and a fixed price or a capped estimate.

  2. 02

    Design

    Architecture, data flow and the evaluation criteria we will be judged against. Agreed in writing before code is written.

  3. 03

    Build

    Short iterations with weekly demos. Everything lives in your repository from day one.

  4. 04

    Validate

    Tests, benchmarks and evaluations against the criteria from step two. Nothing ships on a feeling.

  5. 05

    Handover

    Runbooks, training and a support window. You own the code, the models and the data.

Read the working principles →

05 / Engagement

Three ways to work together.

Project

Fixed scope, fixed price, defined deliverables. Best for a dataset to analyse, a pipeline to build, or a model to ship.

Typically 2 to 12 weeks

Embedded

One senior bioinformatician on a monthly retainer, in your Slack and your repository, with weekly planning.

Monthly, cancel any time

Advisory & training

Architecture reviews, pipeline audits and hands-on workshops on Nextflow, single-cell analysis, agents and MLX.

Half-day to one week

Correspondence

Write to arunbodd@primeomicx.com

Send a few lines about what you are working on. You will hear back within two working days.

Start a project