04 / 08 · /work

What I've actually shipped.

Eight case studies. The flagships first: AGF AWS infrastructure, agflive.com, the Co-Scientist hypothesis engine, and EnzymeForge. The rest below. Stack and impact for each, plain-English.

all repos → ↓ CV

// flagship work

The work I'd lead with.

  1. 2025 – ongoing · Lead engineer · architect

    AWS LambdaDynamoDBS3EventBridge +4

    AGF AWS Infrastructure

    Serverless data pipeline ingesting metadata from 31 laboratory instruments via Lambda → DynamoDB → S3 → EventBridge. Pairs with the AGF Data Sync CLI running on each instrument PC: file discovery, SHA-256 dedup, metadata extraction, staff attribution, experiment detection.

    Impact: From raw instrument output to queryable scientific data, and back to the LIMS the scientists actually use. Infrastructure-as-code, role-based access via Cognito, deployed by CloudFormation. The substrate the rest of the AGF stack runs on.

    → agflive.com→ Private repo · AWS infra
    ▸ Finance dashboard preview
    AGF Finance Dashboard mockup placeholder

    AGF Finance Dashboard: 8+ interactive HTML viz types, forecasting, duplicate detection, GUI for non-technical users.

  2. 2025 · Designer + builder · solo

    Next.jsTypeScriptTailwindAWS Amplify +2

    agflive.com: public website + internal dashboard

    Public face of the Australian Genome Foundry. Server-rendered Next.js, Cognito auth with role-based access, GitHub-based CMS, equipment catalog for 31 instruments.

    Impact: Serves the national facility's public presence. Designed and built solo end-to-end from brief to deployed; sponsor surface for NCRIS / Bioplatforms Australia / NSW Government / Macquarie.

    → agflive.com

  3. 2026 · Solo builder · ~54k LOC

    LangGraphFastAPIReactChromaDB +3

    AGF Co-Scientist: 10-agent hypothesis engine

    A 10-agent AI Co-Scientist: ten specialised LLM agents debating hypotheses via Elo-tournament ranking, grounded in live literature search (PubMed, Semantic Scholar, Tavily). ~54k lines.

    Impact: Multi-agent orchestration, structured debate mechanics, and literature-grounded reasoning at production scale. Pluggable multi-LLM backend (Gemini / OpenAI).

    → github.com/SynBioExplorer/AGF_Co-Scientist

  4. 2026 · Solo builder · research pipeline

    RFdiffusion3LigandMPNNESM-2Boltz-2 +4

    EnzymeForge: computational enzyme design

    A GPU-accelerated 14-stage enzyme-design pipeline for metabolic-pathway optimisation in S. cerevisiae. From an FBA-guided bottleneck call: RFdiffusion3 backbone diversification, LigandMPNN ligand-aware sequence design, a zero-shot fitness ensemble (TranceptEVE + ESM-2 + GEMME + ThermoMPNN), Boltz-2 holo docking with product-inhibition modelling, optional GaMD sampling, Protenix cross-validation, and yeast codon optimisation with Golden-Gate plate maps.

    Impact: Structure-based, ligand-aware protein design end-to-end on the modern toolchain (RFdiffusion / LigandMPNN / Boltz / AlphaFold). Funnels ~2,000 candidate backbones to 48-96 synthesis-ready DNA sequences per enzyme in 2-3 days on a single GPU.

    → Private repo · overview

// supporting work · 4 more

The rest of the catalog.

  1. 2024 – present · Solo · published · live

    Welly: growth-curve web tool

    Web tool that turns plate-reader CSV/XLSX exports into publication-ready growth-curve plots in under a minute. Computes max growth rate + AUC per well, aggregates replicates, generates interactive heatmaps + line graphs, downloadable reports.

    PythonFlaskPlotlypandasPythonAnywhere
    → Live tool→ github.com/.../Welly

  2. 2024 – 2025 · Pipeline lead

    Synechococcus PCC 11901 transcriptomics

    RNA-seq of fast-growing cyanobacterium Picosynechococcus sp. PCC 11901 under 20 stress conditions across nutrient, environmental, and circadian experiments. 60 samples, 240 FASTQ files, full pipeline QC → trimming → quantification → DE → pathway enrichment.

    SalmonDESeq2gseapyfastpSortMeRNAFastQCMultiQCPlotly
    → github.com/.../Synechococcus_PCC_11901_transcriptomics

  3. 2024 – 2025 · Solo builder

    Basejumper: desktop Nanopore pipeline

    Desktop GUI pipeline for Oxford Nanopore sequencing: basecalling (Dorado), demultiplexing, QC (NanoPlot / Porechop), de novo Flye assembly, reference-based evaluation (QUAST). One-click workflow for the lab.

    JavaScriptElectronDoradoFlyeNanoPlotQUAST
    → github.com/.../basejumper

  4. 2026 · Solo builder

    Claude Code agentic orchestration

    Multi-agent orchestration for Claude Code: planner-architect (Opus) → supervisor (Sonnet) → workers (Sonnet) → verifier (Haiku). Behavioural-enforcement patterns plus contract-versioned file boundaries let many agents work the same repo in parallel without stepping on each other.

    Pythontmuxgit worktreesClaude CodeDAG scheduler
    → github.com/SynBioExplorer/Claude_Code_agentic_coding