LevelUp.ai transforms raw learner profiles into structured NSQF Level 2–6 career pathways using a swarm of specialized AI agents — connecting India's 400M+ vocational learners to education, skills, and employment.
LevelUp.ai was built at ImpactX Hackathon to address one of India's most persistent structural challenges — the disconnect between vocational training and meaningful employment.
The platform acts as a personal CareerOS: it understands a learner's current position, maps the NSQF-aligned journey to their goal, and guides them through every step with AI-curated content, skill verification, and job matching.
| Target Users | India's 400M+ vocational learners in Tier 2 & 3 cities |
| Framework | National Skills Qualification Framework (NSQF), Levels 2–6 |
| Architecture | PWA-ready · Mobile-first · Offline-capable |
| AI Backbone | 7 specialized LangGraph agents (GPT-4o, Gemini, Groq) |
| Credentials | Blockchain-verified NSQF certificates on completion |
The Path Builder Agent takes a learner's raw profile — role preference, quiz scores, resume data, and GitHub language history — and constructs a personalized NSQF Level 2–6 roadmap using LangGraph. Each node in the roadmap contains a level title, curated topic list, duration estimate, and completion status. The pathway is persisted to Supabase and cached in LocalStorage for offline access.
The roadmap is rendered as a live, drag-and-drop React Flow DAG (Directed Acyclic Graph). Each node is a custom PathwayNode component showing level, status (completed / in-progress / locked), topics, and duration. Animated edges indicate completion state. Clicking an unlocked node navigates to the Learning module for that level.
For every topic in the roadmap, the Learning Curator agent fetches a verified YouTube tutorial. It uses a hand-curated video library with fuzzy keyword matching as the primary lookup, falling back to a Groq Llama 3.3-70B call for topics not in the library, and a domain-based heuristic as a final fallback. This ensures learners always get a real, working video — never a broken link.
The Quiz Master agent runs a short adaptive quiz during onboarding to assess the learner's existing knowledge. Results feed directly into the Path Builder to calibrate the starting NSQF level, skipping foundations the learner has already mastered and surfacing the right entry point for their journey.
At each NSQF level, the Project Generator agent produces a real-world demo task scoped to the learner's role and level. Completing and submitting the demo triggers an automated evaluation; a passing score marks the skill as verified and contributes to unlocking the next level.
The ATS Checker scores an uploaded resume against a target job description — extracting matched keywords, identifying gaps, and returning an ATS compatibility score. The ATS Resume Builder then uses this analysis along with the learner's verified skills and projects to generate a fully formatted, ATS-optimized PDF resume via a headless Puppeteer renderer.
A conversational agent powered by Llama 3 + Composio tool calls helps learners discover relevant jobs, internships, and apprenticeships. Learners describe what they are looking for in natural language; the agent translates intent into structured search queries, fetches live listings, and presents ranked results with apply links.
Every meaningful action — generating a roadmap, verifying a skill, earning a certificate, submitting an application — is automatically recorded in the learner's Memory Timeline. This RAG-powered persistent log provides a chronological view of the entire career journey and drives the certificate panel, where learners can preview and share their earned NSQF badges.
On completing all topics and demos for an NSQF level, a blockchain-verifiable certificate is issued. The certificate card displays the learner's name, completed level, sector, verified skills, and a unique certificate ID. It can be downloaded as a PDF or shared via a public verification link (/verify/:certificateId).
Learners can connect their GitHub account to automatically enrich their profile. The integration analyzes repository language distribution, commit activity, and project names to infer existing technical skills, which are then factored into the initial NSQF level calibration and surfaced on the public portfolio page.
The dashboard aggregates data across the learner's entire journey — NSQF levels completed, skills verified, projects submitted, applications sent, and certificates earned. Recharts visualizations show progress over time and compare the learner's skill profile against current market demand signals.
- Leaderboard — Gamified ranking of learners by verified skill count, certificates, and project scores. Encourages healthy competition within the community.
- Showcase — Public-facing portfolio pages where learners can display verified credentials, projects, and certificates to potential employers.
- Build Together — A collaborative workspace allowing peer learners to form study groups, share resources, and co-work on demo projects.
Layer responsibilities:
- Client — Next.js App Router pages with React Server Components, Framer Motion animations, and React Flow visualizations.
- API — Serverless route handlers that orchestrate AI agent calls, validate schemas with Zod, and interface with the data layer.
- AI Engine — LangGraph multi-agent swarm. Each agent has a single, well-defined responsibility with fallback chains.
- Data — Supabase PostgreSQL for persistence; LocalStorage for offline-first caching on low-bandwidth devices.
- External — YouTube (learning content), GitHub (profile enrichment), Composio (job tool augmentation).
LevelUp.ai uses a multi-agent orchestration pattern. Each agent is independently deployable, scoped to one task, and connected via a typed LangGraph state machine.
flowchart LR
INPUT([User Profile\n+ Target Role])
subgraph SWARM[" LangGraph Agent Swarm "]
QMA["Quiz Master Agent\nGPT-4o · ~2s\nAdaptive skill assessment"]
PBA["Path Builder Agent\nGPT-4o / Gemini · ~3s\nNSQF L2–L6 roadmap"]
LCA["Learning Curator\nGroq Llama 3.3-70B · ~0.5s\nVerified YouTube tutorials"]
GPA["Project Generator\nGPT-4o · ~4s\nReal-world demo tasks"]
ATSC["ATS Checker\nGPT-4o · ~2s\nResume–JD match score"]
ATSR["ATS Resume Agent\nGPT-4o + Puppeteer · ~8s\nOptimized PDF export"]
CMA["Chat Match Agent\nLlama 3 + Composio · ~1s\nJob opportunity matching"]
end
INPUT --> QMA
QMA --> PBA
PBA --> LCA
PBA --> GPA
GPA --> ATSC
ATSC --> ATSR
INPUT --> CMA
style SWARM fill:#050d1a,stroke:#4f46e5,stroke-width:2px,color:#e0e7ff
style QMA fill:#0c2340,stroke:#38bdf8,color:#e0e7ff
style PBA fill:#1e1b4b,stroke:#818cf8,color:#e0e7ff
style LCA fill:#0a2b1a,stroke:#34d399,color:#e0e7ff
style GPA fill:#1a1040,stroke:#a78bfa,color:#e0e7ff
style ATSC fill:#1a2e0a,stroke:#86efac,color:#e0e7ff
style ATSR fill:#2d1515,stroke:#fca5a5,color:#e0e7ff
style CMA fill:#1a1a0a,stroke:#fcd34d,color:#e0e7ff
| Agent | Model | Latency | Input | Output |
|---|---|---|---|---|
| Path Builder | GPT-4o / Gemini | ~3s | Profile + target role | NSQF L2–L6 node array |
| Learning Curator | Groq Llama 3.3-70B | ~0.5s | Topic + role | { videoId, title, channel } |
| ATS Resume | GPT-4o + Puppeteer | ~8s | User profile + JD | PDF (base64) |
| ATS Checker | GPT-4o | ~2s | Resume + JD | Score + keyword gaps |
| Chat Match | Llama 3 + Composio | ~1s | User intent | Job listings |
| Quiz Master | GPT-4o | ~2s | Onboarding answers | Skill snapshot |
| Project Generator | GPT-4o | ~4s | NSQF level + role | Project spec |
levelup-ai/
├── app/
│ ├── api/
│ │ └── agents/
│ │ ├── path-builder/ # NSQF pathway generator
│ │ ├── learning-curator/ # YouTube video curator (Groq)
│ │ ├── ats-resume/ # AI PDF resume generator
│ │ ├── ats-checker/ # Resume–JD scorer
│ │ ├── chat-match/ # Conversational job matching
│ │ ├── quiz-master/ # Adaptive skill assessment
│ │ └── generate-projects/ # Demo project generator
│ │
│ ├── pathway/ # React Flow roadmap visualization
│ ├── learning/ # Course & module player
│ ├── memory/ # RAG-backed career timeline
│ ├── opportunities/ # Job listings & apprenticeships
│ ├── resume/ # Resume parsing & ATS builder
│ ├── dashboard/ # Analytics & progress tracking
│ ├── github/ # GitHub integration UI
│ ├── leaderboard/ # Community skill showcase
│ ├── buildtogether/ # Peer collaboration workspace
│ ├── portfolio/ # Auto-generated learner portfolio
│ ├── showcase/ # Public skill verification
│ ├── verify/ # Certificate verification portal
│ ├── pulse/ # Live skill pulse feed
│ ├── profile/ # Learner profile management
│ ├── auth/ # Authentication flows
│ └── settings/ # User preferences
│
├── components/
│ ├── landingpage/ # Landing page sections
│ ├── certificate/ # Achievement modal
│ ├── roadmap/ # Custom React Flow node types
│ ├── memory/ # Timeline UI components
│ ├── providers/ # React context providers
│ └── ui/ # shadcn/ui atomic components
│
├── lib/
│ ├── api-client.ts # API base URL configuration
│ ├── userStore.ts # Supabase CRUD helpers
│ └── supabase.ts # Supabase client initialization
│
├── types/ # TypeScript interfaces & schemas
├── backend/ # Python auxiliary backend
├── scripts/ # Database setup scripts
├── public/ # Static assets
├── .env.example # Environment variable template
└── next.config.ts # Next.js configuration
| Technology | Version | Role |
|---|---|---|
| Next.js | 16.1.1 | App Router, SSR, PWA host |
| React | 19.2.3 | UI component layer |
| TypeScript | 5.x | Static type safety |
| Tailwind CSS | v4 | Utility-first styling |
| shadcn/ui | Latest | Accessible component primitives |
| Framer Motion | 12.x | Animations & page transitions |
| React Flow | 11.x | Interactive roadmap DAG |
| Recharts | 3.x | Data visualization |
| Three.js / R3F | 0.184 | 3D background visuals |
| Lucide React | 0.562 | Icon library |
| Technology | Role |
|---|---|
| LangChain + LangGraph | Multi-agent state machine orchestration |
| OpenAI GPT-4o | Pathway generation, resume building, assessment |
| Google Gemini | Fallback LLM for pathway tasks |
| Groq Llama 3.3-70B | Ultra-fast learning content curation |
| Composio | Tool-augmented agent actions (job search) |
| Vercel AI SDK | Streaming AI responses to the client |
| Zod | Schema validation for all agent outputs |
| Technology | Role |
|---|---|
| Supabase | PostgreSQL database + real-time subscriptions |
| Vercel Postgres | Serverless SQL for analytics workloads |
| Firebase | User authentication & account management |
| NextAuth v5 | Server-side session management |
| Puppeteer | Headless Chrome for PDF resume rendering |
| pdf-parse | Resume PDF text extraction |
| Cheerio | Web scraping for job opportunity aggregation |
- Node.js ≥ 20.x
- npm ≥ 10.x
- Git
git clone https://github.com/Dakshin10/LevelUp-ai.git
cd LevelUp-ai
npm installcp .env.example .envOpen .env and populate the following:
# Supabase
NEXT_PUBLIC_SUPABASE_URL=
NEXT_PUBLIC_SUPABASE_ANON_KEY=
SUPABASE_SERVICE_ROLE_KEY=
# Database
POSTGRES_URL=
# AI Models
OPENAI_API_KEY=
GOOGLE_GENERATIVE_AI_API_KEY=
GROQ_API_KEY=
CURATOR_GROQ_KEY=
# Authentication
AUTH_SECRET=
NEXT_PUBLIC_FIREBASE_API_KEY=
NEXT_PUBLIC_FIREBASE_PROJECT_ID=
NEXT_PUBLIC_FIREBASE_AUTH_DOMAIN=
# Integrations
GITHUB_TOKEN=
COMPOSIO_API_KEY=npm run setup-dbThis runs the schema migrations against your Supabase project.
npm run devThe application will be available at http://localhost:3000.
npm run build
npm run starttimeline
title LevelUp.ai — Roadmap to Scale
section Q3 2026
ONDC Integration Protocol : Plug into the Open Network for Digital Commerce
: Exchange training vouchers and credentials programmatically
: Enable regional skill centers to participate
section Q4 2026
Offline Conversational Channels : SMS and WhatsApp career agents
: Operate without an active data plan
: Ultra-low latency model inference
section Q1 2027
Government Multi-Tenant Portals : Dashboards for state skill development corporations
: District-level skill shortage heatmaps
: Localized vocational metrics and job alignment
section Q2 2027
W3C Verifiable Credentials : Cryptographically signed NSQF certificates
: Employer-verifiable portfolio badges
: Instant background validation via DID
| Integration | Purpose | Target |
|---|---|---|
| NCVET API | Official NSQF qualification validation | Q3 2026 |
| WhatsApp Bot | Career guidance for low-bandwidth users | Q4 2026 |
| Vernacular NLP | Support for Hindi, Tamil, Telugu, Kannada | Q4 2026 |
| DigiLocker | Government-backed credential storage | Q1 2027 |
| W3C Verifiable Credentials | Decentralized, employer-verifiable badges | Q2 2027 |
| ONDC Plugin | Skill voucher marketplace integration | Q3 2027 |
Built by Team: ImpactX:
|
|
|
This project is licensed under the MIT License. See LICENSE for details.
Made for Bharat's next billion learners.
