Building PathFinder AI: A Career Discovery Companion Powered by Amazon Nova
How I used Amazon Nova Lite's conversational and multimodal capabilities to build an AI career companion that helps anyone discover their passions and own their path forward.
Many communities lack access to career guidance. Kids grow up not knowing what's out there, carrying pressures that leave no space to discover their passions. The programs that do exist often end, and people are left without tools to continue their growth.
I built PathFinder AI as a personal project to address this gap. It is an AI-powered career discovery and development companion that guides anyone through real conversation — not forms or quizzes.
This post walks through how I built it using Amazon Nova Lite and a serverless AWS architecture.
What PathFinder AI does
PathFinder AI takes a user from zero career knowledge to having a professional profile, a tailored CV, and matched job listings. The entire experience happens through natural conversation.
The app has six core capabilities:
- Discovery conversation — Amazon Nova Lite asks what excites you, what you are good at, and what kind of work makes you feel alive.
- Multimodal certificate analysis — Upload a certificate image or PDF and Amazon Nova Lite extracts the credential name, issuer, skills, and dates automatically.
- Profile generation — Amazon Nova Lite synthesizes everything from the conversation into a structured professional profile.
- CV generation — A full CV built from conversation data, with the ability to tailor it for specific job postings.
- Journey tracking — Goals, achievements, and challenges detected automatically from the conversation.
- Job matching — Real job listings matched to the user's skills and career direction.
It works for any field. Tech, sports, trades, beauty, music, healthcare — whatever drives the user.
Why Amazon Nova Lite
I chose Amazon Nova Lite for three reasons.
First, multi-turn conversation with context. PathFinder AI runs a 4-phase conversation engine: Discovery, Skills Assessment, Path Generation, and Active Coaching. Amazon Nova Lite maintains context across long conversations and handles state transitions naturally. The user does not feel like they are being routed through a workflow — it feels like talking to a mentor.
Second, multimodal understanding. Certificate analysis was a requirement from day one. Users in underserved communities often have credentials but no idea how to translate them into a resume. Amazon Nova Lite can read images and PDFs and extract structured data, which made this possible without building a separate OCR pipeline.
Third, structured output generation. Profiles and CVs require specific formats. Amazon Nova Lite reliably generates structured JSON from unstructured conversation data, which the app renders into professional documents.
All AI interactions go through the Amazon Bedrock Converse API. This provided a consistent interface for conversational AI, document analysis, and content generation without managing multiple model integrations.
Architecture
The stack is fully serverless on AWS:
- Frontend: React, TypeScript, and Tailwind CSS hosted on Amazon S3 behind Amazon CloudFront
- API: Amazon API Gateway (REST) with Amazon Cognito authorization
- Backend: Six AWS Lambda functions in Python covering message handling, state management, portfolio (profile and CV), journey tracking, opportunities, and admin
- AI: Amazon Nova Lite via Amazon Bedrock Converse API
- Storage: Amazon S3 for user data and uploaded certificates, Amazon DynamoDB for metadata
Each Lambda function handles a specific domain. The message handler manages the conversation engine and certificate uploads. The portfolio handler generates profiles and CVs. The journey handler tracks goals and challenges extracted from conversation. The opportunity handler matches job listings to user profiles.
How the conversation engine works
The conversation engine is the core of PathFinder AI. It manages four phases:
Phase 1 — Discovery. Amazon Nova Lite asks open-ended questions about passions, interests, and aspirations. It reflects back what it hears and digs deeper. After several exchanges, it builds a picture of who the user is.
Phase 2 — Skills Assessment. Amazon Nova Lite guides the user through capturing credentials. This is where multimodal analysis comes in. The user uploads a certificate image or PDF, and Amazon Nova Lite extracts the details. The conversation continues naturally around what the credential means to the user.
Phase 3 — Path Generation. Based on discovered passions and validated skills, Amazon Nova Lite generates a personalized learning path with concrete next steps.
Phase 4 — Active Coaching. Ongoing career coaching where the user can ask to build their profile, generate a CV, set goals, or find job opportunities. Everything feeds back into their profile automatically.
State transitions happen organically. Amazon Nova Lite decides when the user is ready to move forward based on conversation depth, not a fixed number of messages.
Multimodal certificate analysis
This feature was one of the most impactful to build. Here is how it works:
- The user uploads a certificate image (JPG, PNG) or PDF through the chat interface.
- The frontend sends the file to the upload Lambda, which stores it in Amazon S3.
- The message handler retrieves the file and sends it to Amazon Nova Lite via the Converse API with a multimodal prompt.
- Amazon Nova Lite analyzes the document and returns structured data: credential name, issuing organization, skills validated, and date earned.
- The extracted data is stored in the user's profile and confirmed back in conversation.
For PDFs, the Lambda converts pages to images before sending to Amazon Nova Lite. This handles multi-page certificates and varied PDF formats.
The result: a user who just earned their first qualification can upload a photo of it and immediately see it reflected in their professional profile. No data entry required.
Journey auto-tracking
Journey tracking works by analyzing conversation content after each exchange. When a user mentions a goal, a challenge, or an achievement, Amazon Nova Lite detects it and logs it to the journey tracker.
This happens transparently. The user does not fill out forms or click buttons to track their progress. They just talk, and PathFinder AI keeps track.
Challenges
Conversation state management. Maintaining coherent multi-turn conversations across Lambda invocations required careful state design. Each conversation turn loads the full history from Amazon S3, sends it to Amazon Nova Lite, and persists the updated state.
Multimodal reliability. Certificate images vary in quality, orientation, and layout. Tuning prompts to handle edge cases like blurry photos, certificates in different languages, and PDFs with complex formatting took iteration.
Natural phase transitions. Early versions felt robotic when transitioning between conversation phases. The solution was giving Amazon Nova Lite explicit instructions about transition criteria while letting it choose the natural moment to move forward.
What I learned
Amazon Nova Lite is capable at maintaining context across long multi-turn conversations. The Converse API makes it straightforward to build structured AI interactions with tool use. Multimodal understanding opens up possibilities that text-only models cannot match — certificate analysis was a key differentiator. Building for communities that lack career guidance requires a fundamentally different approach than traditional career tools.
What is next for PathFinder AI
The roadmap includes WhatsApp integration to meet users where they are, universal job matching across all industries, real-time course APIs for verified learning resources, local event matching, and multi-language support.
The technology works. The next step is getting it into the hands of communities that need it most.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article