Skill Path Finder

Inspiration

Growing up in India, we saw brilliant young people from small towns — full of potential, full of dreams — but completely lost when it came to their careers. They didn't know what to learn, couldn't afford a career counselor, and many didn't even have stable internet. A single counseling session costs ₹500 to ₹2000 — something most families simply cannot afford. We felt that was deeply wrong. That frustration became our motivation — and SkillPath Finder became our answer.

What It Does

SkillPath Finder is an AI-powered career counseling platform that gives every Indian a free personalized career roadmap in under 60 seconds.

At its core is Priya — our AI career counselor powered by Amazon Bedrock (Claude 3 Haiku). She asks 6 simple questions — name, interest, specific goal, experience level, time availability, and city. Based on the answers, she instantly generates a complete 3-month career roadmap with free YouTube resources in Hindi, government course recommendations like PM Kaushal Vikas and Skill India, real job listings matched to their city, expected salary ranges, and a Skill Passport that tracks learning progress with digital credentials.

What makes SkillPath Finder truly unique is how it reaches users — three channels, every Indian covered:

🌐 Web App — full experience on any browser
💬 WhatsApp — works on basic 2G, just send "Hi"
📞 Voice Call — call from any phone, Priya speaks to you, roadmap arrives by SMS, zero internet needed

Our Story

We didn't build this for a hackathon. We built this because we genuinely felt the problem.

We have seen friends and family members waste years going in the wrong direction — studying the wrong things, missing opportunities — simply because nobody ever sat them down and said "here is what you should learn and here is how to get there."

When we discovered Amazon Bedrock and realised we could build a personalised AI counselor that responds in seconds — we knew exactly what we wanted to build. Something that felt like talking to a real mentor. Something that worked on WhatsApp and a basic phone call — not just on a laptop in a city. Something that reached the people everyone else forgot to build for.

That became Priya. That became SkillPath Finder.

What We Built

We built SkillPath Finder as a fully serverless platform on AWS with three complete access channels.

The web app is built with React.js and Vite, hosted on AWS Amplify with automatic CI/CD from GitHub on every push. The backend runs entirely on AWS Lambda (Python 3.14) connected through Amazon API Gateway, with Amazon DynamoDB storing all user profiles, sessions, and roadmap data.

The AI layer uses Amazon Bedrock (Claude 3 Haiku) for real-time roadmap generation — analysing each user's unique inputs across 7 career branches and 15 personalised paths. We also built a smart fallback system that generates quality roadmaps when Bedrock is unavailable, ensuring 100% response rate always.

The WhatsApp bot connects Twilio's WhatsApp Business API to AWS Lambda — users send "Hi" and get a full AI counseling conversation in their chat. The voice call channel uses Twilio Voice and TwiML connected to a separate Lambda in us-east-1 — Priya speaks, the user answers, and an SMS with the roadmap arrives the moment the call ends. All sessions are stored in DynamoDB with in-memory fallback for maximum reliability.

Challenges We Faced

Nothing about this was straightforward.

Amazon Connect was completely blocked. Our original plan was to use Amazon Connect for the voice call feature. After hours of trying, we discovered that AISPL accounts cannot create Amazon Connect instances anywhere. We had to pivot entirely to Twilio Voice and TwiML — rebuilding the voice flow from scratch.

WhatsApp messages were arriving empty. Every message Twilio sent to our Lambda webhook showed up as an empty body. After long debugging sessions we realised Twilio sends Base64 encoded request bodies — and we weren't decoding them. One fix, but hours to find.

DynamoDB region mismatch. Our main stack runs in ap-south-1 but Twilio routes better through us-east-1 — causing AccessDeniedException errors across our Lambdas as they looked for DynamoDB in the wrong region.

Bedrock payment restrictions. Our AISPL account hit an AccessDeniedException: INVALID_PAYMENT_INSTRUMENT error due to UPI AutoPay restrictions. We built a complete fallback roadmap generation system so users always receive a quality response regardless of Bedrock availability.

Duplicate WhatsApp messages. After fixing Base64, users were receiving 3 messages at the end of every conversation — the thinking message, the roadmap, and an error message. We refactored the Lambda to return None when the location stage handled sending directly, and added a None check before any send call in lambda_handler.

Every single one of these challenges made the final product stronger. And honestly — solving each one taught us more than months of regular learning ever could.