Top Line View Advisor

Advisor is your pocket CTO. We help small businesses ditch expensive consulting fees, using conversational AI to identify technology risks and opportunities, and deliver tailored, actionable plans.

Comment

Inspiration

Throughout my career as a technology leader at high-growth scaleups, one of my core responsibilities has always been translating complex technical concepts into clear, actionable decisions for non-technical executives.

Most recently, as CTO of an eLearning company, I led the development of a catalogue of over 150 digital skills lessons, covering Ad Tech, MarTech, Data & Analytics, ultimately growing it to a successful exit, acquired by QA, the UK's largest technology training provider, in 2021.

Early this year, I began exploring the capabilities of local AI agents and agentic harness architectures. When the Gemini XPrize competition was announced, a global challenge to apply AI to meaningful real-world problems, I spent weeks stress-testing different product ideas.

The opportunity became clear: combine my experience in technology leadership and digital education to build an intelligent advisory platform that scales CTO-level thinking to small businesses, the very businesses that have never been able to afford it.

I'm also acutely aware that we're at an inflection point for small businesses. AI and automation now put capabilities once reserved for large enterprises such as intelligent data analysis, automated workflows, personalised customer engagement, within reach of even the smallest teams. But that same wave of innovation has dramatically expanded the attack surface for cybercrime. SMBs are increasingly targeted precisely because they lack the defences of larger organisations. The opportunity and the risk are two sides of the same coin, and TLV Advisor is built to help businesses seize one without being blindsided by the other.

What does Top Line View Advisor do?

Top Line View (TLV) Advisor is an outcome-focused, "CTO-as-a-Service" platform designed specifically for soletraders and small businesses. Instead of rushing to focus on the latest technologies it acts as a trusted, neutral advisor that diagnoses a business's operational health across 10 core pillars (like Sales, Data, Operations, and Cybersecurity) to ensure the business has some solid foundations to leverage the latest technologies.

Moving away from overwhelming static forms, TLV Advisor uses a conversational, iterative audit to assess the business's technology maturity level. It then generates simple, actionable, and prioritised remediation plans to fix actual business bottlenecks, eliminate wasted software spend, and ensure businesses build a secure data foundation before attempting to deploy advanced AI.

Crucially, TLV Advisor isn't a one-off audit, it is built with persistent memory so its knowledge of the business compounds over time. Moving away from overwhelming static forms, it uses a conversational, iterative audit that provides seamless continuity across sessions. After generating an initial improvement plan to take businesses to a base line, the Advisor proactively reviews each business area on a rolling three-month cycle. This transforms tech strategy from a forgotten presentation into a powerful, continuous improvement programme.

How we built it

I built a highly modular, cross-platform architecture adhering to modern 12-factor app principles using Google Antigravity.

Frontend: A responsive Flutter application compiled for both Web (acting as the top-of-funnel lead generation) and Native targets. Backend & Auth: Firebase Authentication handles secure access, while Cloud Firestore (NoSQL) maintains the conversational memory state, user maturity profiles, and active remediation plans. AI Middle-Tier: We deployed Python Gen-2 Cloud Functions as a secure API middleman to handle all LLM orchestration. Multi-Model Routing: We utilised the Google GenAI SDK to route requests dynamically: Gemini Flash handles the fast, conversational bite-sized questioning, while Gemini Pro handles the heavy computation required to generate the comprehensive, structured project plans.

It is built with privacy in mind with all data being retained in the product tenancy.

Challenges we ran into

  • Regional AI Availability & Latency: Navigating the regional availability of the latest GenAI models while balancing cost and latency. Currently, our secure Cloud Functions are hosted in Belgium, but they are orchestrating calls to the 2nd generation Gemini models hosted in the Central US available through Vertex.
  • Wrangling AI Coding Assistants: While AI assistants are powerful, getting them to strictly adhere to a project's architectural vision is difficult. Even with extensive AGENTS rules, workflows, and explicit skills, they require constant vigilance to prevent them from breaking modularity, hallucinating dependencies, or attempting to read sensitive secret files.
  • Stateful Conversational Audits: [Proposed] Designing an AI that doesn't just chat, but conducts a structured interview. It was challenging to engineer prompts that keep the LLM focused on extracting the user's maturity level, saving that state accurately to Firestore, and knowing exactly when to pivot from "asking questions" to "generating a plan."
  • Parsing LLM Output for UI: [Proposed] Getting the Gemini Pro model to consistently return complex remediation plans in a strict, predictable JSON format so the Flutter UI could confidently parse and render them into interactive checklist items without breaking.

Accomplishments that we're proud of

  • Mastering a New Framework on the go: Building my first-ever Flutter application for this project was a massive, fast-paced education. I learned that while AI agents are incredible accelerators for generating UI and translating concepts, you still have to step back and deeply understand Flutter’s underlying state management and widget life cycles in order to architect the app properly and avoid potential complex cross-platform issues.
  • Frictionless Onboarding: I am incredibly proud of the onboarding flow. By tracking device intelligence passively and asking limited questions but enough to tailor the experience rather than create a massive friction up front.
  • The Maturity Matrix Engine: Developing a robust, scalable Maturity Matrix across 10 core business pillars. It took significant effort to design a matrix that accurately reflects real-world SMB capabilities and can be continually improved over time without breaking legacy user audits.
  • Multi-Model Efficiency: Successfully implementing the dual-model routing strategy. Using Gemini Flash for the chat keeps the experience snappy and costs low, while reserving Gemini Pro only for the heavy, high-value plan generation.
  • UI as an AI Educator (Implicit Learning): I am incredibly proud of designing an interface that doesn't just provide answers, but actively teaches the user about AI's capabilities. The UI intentionally exposes different "thinking styles," different models, and visibly demonstrates how the AI leverages "memory" across sessions. Simply by interacting with the platform, small business owners organically improve their underlying AI literacy and begin to truly understand what the technology is capable of and will be useful when we roll out future elearning to support more complex initiatives

What we learned

  • The Reality of AI Code Generation: Code assistants help immensely in accelerating development with new technologies (like Flutter and Dart), but they do not replace the need for deep architectural planning. It takes significant time and code-review effort to ensure the model has implemented the logic exactly as specified without introducing silent regressions.
  • Scope Management & "Simplicity First": I originally hoped to weave interactive microlearning modules directly into the AI's project plans. However, given the hackathon time constraints, I had to ruthlessly prioritise the core audit loop. It reinforced the lesson that shipping a complete core feature is better than shipping multiple half-finished ideas.
  • Prompt Engineering is Software Engineering: I learned that prompting isn't just typing English; it requires the same rigorous testing, version control, and edge-case handling as traditional code, especially when relying on the LLM to trigger database saves and state changes.

What's next for TLV Advisor

  • GTM Feedback: Launch to friends and family in target market and gather feedback
  • Product Launch: Rolling out our three-tier subscription model with a two-week free trial.
  • Linkedin Test campaign: Executing our Go-To-Market strategy targeting SMB sector professionals most likely to be active on LinkedIn and measuring conversion rates. Updating product on customer feedback from trials.
  • Platform Expansion: Official release of the dedicated mobile applications and the installable PWA to increase user retention and session frequency.
  • The Audit & Assessment Flywheels: Implementing a continual improvement program for the underlying Maturity Matrix, driven directly by the aggregated, anonymised data from user assessments.
  • SME Technology Radar: Introducing the deferred microlearning features—providing businesses with an "Emerging Technology Radar" to educate them on relevant innovations before they become critical threats or missed opportunities.

Built With

Share this project:

Updates