How to Rapidly Build GenAI Proof of Concepts for Delivery & Scaling

How to Rapidly Build GenAI Proof of Concepts for Delivery & Scaling

This blog is about quick and easy steps you can take to embrace AI. In just a few clicks, you'll be creating solutions to high-impact problems that were previously out of reach.

Step 1: Define the challenge to solve. As a Technology Consultant for the Public Sector, I have defined two challenges that can be rapidly prototyped using AI.

• Front-end journeys where people apply, pay and receive updates can be frustrating and disconnected. Examples include applying for passports, requesting information or reporting life events.
• Case Workers such as Probation Officers, Social Workers and Mental Health Assessors spend a lot of time reviewing long assessment reports and case notes to identify key points and find critical information.

Step 2: Use Chat-GPT to help define the problem. In your prompt, describe the challenge, ask for areas of consideration, pain points for key personas and possible solutions. This will give you a full idea canvas in a single reply.

Step 3: Select the most promising pattern from the list and tell Chat-GPT you'd like to build an app to address this pain point. Request that it asks YOU questions to further clarify how the app should work, and with that information, ask Chat-GPT to create user stories for the personas in Behaviour Driven Development (BDD) format. BDD is a way to write requirements and acceptance criteria that uses human-readable language to define how a system should behave focusing on examples and scenarios.

Step 4: Use Chat-GPT to define your working prototype. AI Generators only need a clear story, not code. Tell Chat-GPT to make a comprehensive prompt to use in your chosen AI Generator to create the app.

Step 5: Use an AI Generator such as Bolt.new, UI Bakery or Windsurf. I use Bolt, which is a browser-based platform for building and deploying full stack web applications from natural language prompts. Paste the prompt in the free text field and within three minutes Bolt will create a preview of a working application.

Personal Experience

For my front-end journeys use case described above, Chat-GPT identified some really intuitive observations such as fragmented journeys and high citizen expectations. I asked for solutions in a tabular format and Chat-GPT provided nine solutions including a Unified Citizen Portal, Conversational AI, and Guided Help.

I focused on the issue of confusing, non-tailored citizen journeys and asked Chat-GPT to help me clarify app functionality needed. It responded with questions about journey design, personalisation, integration, and output.

I chose "reporting a birth" as the use case and requested Chat-GPT to draft a detailed prompt for creating the web app in Bolt.

Within minutes, Bolt developed a Birth Reporting app with a GOV.UK-inspired design and proper accessibility considerations. The app accommodates various user types including UK citizens, individuals living abroad, single parents, and couples. It generates tailored checklists with relevant steps and resources. The interface displays priority indicators for checklist items, offers interactive progress tracking and includes links to official government resources.

I applied the same process for the Case Worker scenario, focusing on the Probation Officer Persona, whose pain points are delays and missed red flags due to manual report scanning.

Chat-GPT asked me questions on the functional scope such as the type of documents the app will process, workflow integration, data sensitivity levels, and output formats. It created User Stories in behaviour-driven development (BDD) format for both Probation Officer and Supervisor Personas.

I asked Chat-GPT to generate a prompt for Bolt to build the app, including key features like AI summarisation, risk flagging, a simple interface, relevant personas, and tech stack suggestions.

In just three minutes, Bolt developed "Risk Lens," a comprehensive web application for Probation Officers to efficiently analyse assessment reports and case notes using AI-driven risk identification and summarisation. Notable features include role-based authentication, drag-and-drop document upload, AI-enabled risk scoring, and a real-time dashboard displaying statistics and risk distribution charts.

Expanding Your Prototype to Enterprise Scale

After confirming your prototype solves the challenge, begin rapid development by experimenting, updating code, and deploying with live data according to organisational standards. Your app could be in production in just 12 weeks.

The key is to involve the right decision makers from the start and surround yourself with a team of experts that includes a UX Lead, Data Scientist, Full-Stack Developer, DevOps, and Domain SME. Remember that technical debt slows down AI, so enterprise level adoption also means legacy refactoring and modernisation.

Once the foundations are in place, you can envision a factory-model of rapid builds where you design, test, simulate, extract, prioritise, and operationalise AI-based applications that solve high-impact challenges for your organisation.

Reflection

In essence, Bolt developed two practical demonstrations illustrating the potential for personalised government applications to enhance citizen experience, minimise manual processing time, and highlight essential information, in-line with my brief.

You might have an idea but have been held back because you can't quickly build a website and landing page. You might actually have coding skills but want to speed up your workflow or quickly iterate on ideas with simple prototypes. Or like me, you might be a Technology Consultant who wants to demonstrate the art of the possible and refine requirements using prototypes that can be tailored in real time.

By following these simple steps, you can now create landing pages or build functional apps in your own field of work (or play) without relying on advanced development skills or a significant investment of time.