LabACT

Pharma and biotech engineering

LabACT

AI consulting for systems you can audit

I help regulated teams turn brittle prototypes and manual workflows into software they own. Models draft code and classify documents. Approvals, records, and execution live in your stack.

Probabilistic tools. Deterministic systems.

Get in touchHow I work

Assess

Map the workflow, data, risks, and who owns what

Create

Build deterministic software with AI in the engineering loop

Transfer

Hand off code, docs, controls, and ownership

AI should help build the system, not become it

I use AI across design, build, test, and improvement. What runs in production stays controlled, testable, and owned by your team.

What LabACT means

LabACT is Assess, Create, Transfer. Most engagements stop at delivery. I stay until your team can run and extend the system without me.

  • Assess the work, data, risks, decisions, and constraints.
  • Create deterministic software and controlled AI-assisted workflows.
  • Transfer the code, knowledge, controls, and ownership to you.

A QC group once showed me a ChatGPT workflow that worked in demos but could not pass audit. We rebuilt it as a Django app with signed approvals and a Postgres audit log. Same outcome for reviewers, none of the runtime risk.

Where I sit

Traditional delivery often ignores what AI can speed up. AI-first automation often replaces defined systems with agents that behave differently every run. I work in the gap: fast builds, predictable operations.

  • Use models to investigate, generate code, classify, and summarize
  • Keep rules, records, security, approvals, and execution in software you control
  • Version prompts and models like any other dependency
  • Make important results testable, observable, and recoverable

Three parts of the practice

Consulting, a repeatable build method, and written guidance. All aimed at systems your team can run without standing on prompts or agents.

Embedded consulting

I join teams where a workflow is stalled, fragmented, manual, or stuck inside an unreliable prototype. I work with leadership, domain experts, engineers, and operators to find the real process, remove blockers, build the system, and hand it over.

System-building method

Informal work becomes durable software. AI helps throughout the build. Production runs through code, APIs, databases, queues, schedules, permissions, and explicit state. The worker is software, not a chatbot replaying instructions on every run.

Work descriptionWorkflow specGenerated codeTests and controlsDeployed systemMeasured improvement

Written guidance

Notes for technical leaders, consultants, founders, and operators: which parts of a workflow should use AI, which must stay deterministic, how to turn prompts into maintained software, and how teams keep ownership after outside help leaves.

Four layers

How I structure systems when audit trails, approvals, and data integrity are required.

A typical batch-record review might pull PDFs from a shared drive, run extraction through a model with fixed prompts, write structured fields to Postgres, route exceptions through a Django approval queue, and log every state change. The model helps read documents. It does not decide what ships.

Layer 1

Trusted state

  • Postgres or your system of record
  • Files and APIs
  • Identity and permissions
  • Source-controlled config

Layer 2

Deterministic execution

  • Application code and workflows
  • Validation and routing
  • Schedules and queues
  • Retries and integrations

Layer 3

AI assistance

  • Interpretation and extraction
  • Classification and summarization
  • Code generation
  • Anomaly detection and drafting

Layer 4

Human control

  • Approvals and exceptions
  • Policy decisions
  • Feedback and oversight
  • Final judgment

What I avoid: a prompt that calls an agent that hopes the right answer shows up. Data flows through code. AI sits inside bounded steps. Humans hold accountability where regulations require it.

Four services, one way of working

I use AI to build better software. I do not make AI responsible for running your business.

ACT Assessment

A

Map the current process, tools, data, blockers, risks, and ownership. Decide what to automate, what should use AI, and what stays under human control.

What you get

Technical and operational transformation plan

ACT Build

B

Build the deterministic application, automation, integration, or internal tool. AI speeds up engineering. What ships is normal, maintainable software.

What you get

Working code, infrastructure, tests, documentation, and monitoring

ACT Conversion

C

Take a prompt, agent, prototype, spreadsheet, n8n workflow, or manual process and replace it with a reliable production system.

What you get

Source-controlled, testable, production-grade replacement

ACT Enablement

E

Train internal teams to use AI in engineering without letting operations turn into an uncontrolled runtime.

What you get

Standards, examples, review processes, reusable components, training, and ownership transfer

Why LabACT

Plenty of firms sell AI. Fewer draw a clear line between what models should help with and what software must own.

  • I use AI to build software faster. I do not put models in charge of production behavior.
  • Models belong in the engineering loop. Running systems should behave the same way on Tuesday as they did on Monday.
  • Demos often work once. They fail under repetition, scale, audit, security, maintenance, and handoff.

Lab is where you test. ACT is where you run it for real.

Guides and teardowns

Practical notes for technical leaders, consultants, founders, and operators in pharma, biotech, and regulated work.

Prompt to Production

Turn a useful prompt into code, tests, interfaces, and operations.

Read

The Agent Boundary

Where agents help, where they fail, and when to replace them with software.

Read

Deterministic by Design

Architecture patterns for AI-assisted systems with hard boundaries.

Read

Workflow Teardowns

What belongs in databases, code, AI, and human review.

Read

Own the Worker

Turn repetitive work into client-owned software.

Read

AI Engineering Controls

Versioning, evaluation, approvals, logs, retries, and security gates.

Read

All writing

Johnny Rice

Founder and Principal, LabACT

Embedded AI transformation and engineering enablement

I take unclear workflows and unreliable prototypes and turn them into systems teams can run without outside help.

I work with pharma, biotech, and life sciences teams where audit trails, approvals, data integrity, and operational reliability are required, not optional.

Background in software engineering and operational systems for regulated environments. Recent work includes approval workflows, batch record review tooling, and replacing spreadsheet-and-email processes with owned applications.

Get in touch

A stalled workflow, a brittle prototype, or a transformation effort that needs a concrete system behind it.

hello@labact.com