Fyra — AI Portfolio Assistant
Custom knowledge base assistant that qualifies visitors, recommends projects, and captures leads in real-time.
Problem
Portfolio websites are static. Visitors with specific business goals often leave without knowing if the developer is a match, how much it would cost, or how to submit a structured project brief.
Solution
Built Fyra, a custom-trained AI portfolio assistant. It acts as an interactive client partner: qualifying visitor requirements, recommending matching projects from the portfolio, estimating timelines, and generating structured leads via an embedded form.
Approach
Leveraged Google Gemini API with system instructions for proactive visitor engagement, custom RAG over Jishnu's projects and services, and a reliable lead capture form trigger.
Architecture
Next.js client-side chat interface connected to an API route stream, integrating Supabase database for persistent conversations, and a webhook to deliver lead briefs.
Outcome
- Successfully converted casual visitors to qualified leads by automatically generating project briefs.
- Reduced initial consultation overhead by collecting structured visitor requirements before contact.
- Enabled instant project recommendations based on visitor query matching.
Case Study
Overview
Fyra is an intelligent portfolio assistant integrated into this website. It acts as an autonomous client-onboarding specialist, helping convert website visitors into qualified business leads.
The Challenge
Standard portfolio sites are static lists of links and descriptions. Prospective clients land with specific projects in mind, but must manually read through pages to figure out if the developer has the relevant skills. If they do decide to reach out, the contact process requires back-and-forth emails just to establish the initial scope, budget, and timeline.
Technical Architecture
Fyra is built directly into Jishnu's portfolio:
- Frontend: A custom, polished chat interface with smooth Framer Motion animations and streaming responses.
- AI Core: Powered by Google's Gemini API, utilizing a highly structured system prompt.
- Retrieval Augmented Generation (RAG): A local keyword-matching database query system pulls relevant projects and services based on user queries, providing exact context to the LLM.
- State Integration: Real-time intent detection flags key criteria like budget and timeline. When a lead is ready to submit, it appends a special token to dynamically render a custom contact form in the chat.
Key Features
- Custom Knowledge Base: Seamless access to Jishnu's background, past projects, and technology stack.
- Prompt Engineering: System instructions designed to keep the AI focused exclusively on Jishnu's work and prevent general-purpose prompts or jailbreaks.
- Visitor Qualification: Automatically analyzes if the visitor has a project, budget, or timeline.
- Project Recommendation: Matches user ideas directly to related work Jishnu has done (e.g., smart robotics or mobile apps).
- Lead Capture Workflow: Displays a clean, embedded contact form to collect visitor emails and project summaries.
Outcomes
- Converted passive visitors into structured project briefs.
- Automated client onboarding qualification.
- Zero-dependency, low-latency, and cost-effective local RAG implementation.