*
Career Mentor AI
Powered by Hindsight Memory - Built for HackWithBangalore 2.0
| Event | Organiser | Stack |
|---|---|---|
| HackWithBangalore 2.0 | HackWithIndia | Next.js · Gemini · Hindsight |
01. The Hackathon Journey
HackWithBangalore 2.0, organised by HackWithIndia, brought together some of the region's sharpest developer minds for an intensive coding showdown where innovation met opportunity. The event challenged participants to build meaningful, production-grade solutions under tight time constraints - and our team rose to the occasion with Career Mentor AI.
02. The Problem We Set Out to Solve
Early-career professionals and students face a common and deeply frustrating challenge: navigating the career landscape without a trusted, personalised advisor. Existing solutions fall into one of two categories - generic AI chatbots with no memory of the user, or expensive human mentors that most students cannot access at scale.
The consequences are significant:
- Students repeatedly re-explain their background to AI tools, receiving generic advice that ignores their actual situation.
- Skill gaps go unidentified until it is too late in the application cycle.
- Resume feedback is one-size-fits-all rather than targeted to the individual's experience level and target roles.
- Internship recommendations lack personalisation, leading to poorly matched applications and lower success rates.
Career Mentor AI was designed to address each of these pain points directly by combining the contextual intelligence of a large language model with the persistence of a dedicated memory layer.
03. How It Works - System Architecture
Career Mentor AI is a full-stack web application structured around three distinct layers: the presentation layer, the intelligence layer, and the memory layer. These three work in concert to deliver a coherent, personalised experience that improves with every interaction.
3.1 Presentation Layer
The frontend is built with Next.js 16 and React 19 using TypeScript throughout. The UI is composed of shadcn/ui components styled with Tailwind CSS, delivering a responsive, mobile-first interface with full dark mode support. The design system uses a deep purple primary palette (
Primary: #453f9e | Accent: #b3a6f7 ) that communicates trust and professionalism.
3.2 Intelligence Layer - Google Gemini API
The AI capabilities are powered by Google Gemini, accessed via a custom wrapper in
lib/gemini.ts . Gemini handles four distinct tasks within the application:
- Generating personalised mentor responses in the chat interface.
- Analysing resume content and producing structured, actionable feedback.
- Identifying skill gaps relative to a user-defined target role.
- Scoring and ranking internship matches against a user's stored profile.
All AI calls are mediated through Next.js Server Actions, keeping API keys off the client and ensuring a clean separation between the UI and the intelligence layer.
3.3 Memory Layer - Hindsight (Vectorize)
This is the architectural element that makes Career Mentor AI meaningfully different from a standard AI chat wrapper. Hindsight is a local vector memory system that runs inside Docker and exposes a REST API on
http://localhost:8000 . The application stores three categories of memory:
| Memory Type | What Gets Stored |
|---|---|
| Experiences | Skills learned, projects completed, internship applications submitted, real-world achievements. |
| Observations | Synthesised insights such as 'Strong in frontend, needs backend exposure'; patterns in career interests; areas of strength and weakness. |
| Mental Models | Ideal role type and target company, career aspirations, preferred tech stack, work culture preferences. |
Hindsight's semantic search capability means the AI mentor can retrieve the most contextually relevant memories when composing a response, rather than relying on a simple key-value lookup. The result is advice that feels genuinely informed by the user's history.
04. Technology Stack
The following table summarises the key technologies used across the project:
| Layer | Technology | Purpose |
|---|---|---|
| Frontend | Next.js 16 + React 19 | Full-stack framework with server actions |
| Language | TypeScript | End-to-end type safety |
| Styling | Tailwind CSS + shadcn/ui | Component library and utility-first CSS |
| AI / LLM | Google Gemini API | Natural language generation and analysis |
| Memory | Hindsight (Vectorize) | Persistent semantic memory via Docker |
| Hosting | Vercel | Edge deployment for Next.js |
| Container | Docker + Docker Compose | Local Hindsight service orchestration |
05. Feature Set
Career Mentor AI ships with seven interconnected modules, each backed by the Hindsight memory layer:
Dashboard
A centralised overview of the user's career journey, including a profile completeness tracker, quick statistics on skills and projects, and AI-generated recommended next steps.
Skills Tracker
Users can add and categorise technical and soft skills with proficiency levels. All skills are written to Hindsight and inform every subsequent AI interaction.
Projects Portfolio
Documents completed projects with descriptions, technology lists, and external links. Builds a portfolio foundation that the AI references when generating internship recommendations.
Resume Feedback
Users paste or upload their resume content to receive structured AI feedback with specific improvement suggestions, tailored to their target field and experience level stored in memory.
Skill Gap Analysis
The user specifies a target role or company. The AI cross-references the user's stored skills against inferred requirements and returns a priority-ranked list of missing skills and a suggested learning roadmap.
Internship Recommendations
Generates personalised internship matches with compatibility scores, application tracking, and role-specific tips - all informed by the user's complete stored profile.
Mentor Chat
A real-time conversational interface where the AI mentor draws on the full Hindsight memory context to provide genuinely personalised career guidance. The longer a user engages with the platform, the more contextually accurate the responses become.
06. A Look at the Code
Below is a representative extract showing how the application stores a skill memory and retrieves AI mentor feedback using the Hindsight client and Gemini wrapper:
// Store a skill memory in Hindsight
await storeMemory('skill', {
name: 'React',
level: 'Advanced',
yearsExp: 2,
projects: ['Project A', 'Project B']
});
// Retrieve personalised feedback using stored context
const feedback = await getMentorFeedback(
userId,
'resume',
resumeContent
);
// Generate internship recommendations
const internships = await getInternshipRecommendations(userId);
The Hindsight client in lib/hindsight.ts abstracts all memory read/write operations. When a Gemini call is made, the application performs a semantic search across stored memories to inject the most relevant context into the prompt, resulting in responses that are genuinely tailored to the individual user.
07. Getting Started
To run Career Mentor AI locally, the following prerequisites are required: Node.js 18 or higher, Docker and Docker Compose, and a Google Gemini API key (available free from Google AI Studio).
Installation
# 1. Clone the repository
git clone https://github.com/Apekshakv/hack_with_blr_Career-Mentor-AI
cd career-mentor
# 2. Install dependencies
pnpm install
# 3. Configure environment variables
cp .env.example .env.local
# Add GEMINI_API_KEY to .env.local
# 4. Start the Hindsight memory service
docker-compose up -d
# 5. Start the development server
pnpm dev
# 6. Open http://localhost:3000
Deployment to Vercel
Push the repository to GitHub and import it into Vercel. Set the following environment variables in the Vercel project settings:
- GEMINI_API_KEY - your Google Gemini API key.
- HINDSIGHT_API_URL - the URL of your deployed Hindsight instance.
09. Future Enhancements
The MVP delivered at the hackathon is a solid foundation. The following enhancements are planned for future iterations of the product:
- LinkedIn integration for verified skill import.
- Mock interview module with video recording and AI-evaluated feedback.
- Peer mentoring network connecting students with similar profiles and goals.
- Salary negotiation coaching based on role, location, and experience data.
- Visual career path mapping to illustrate potential progression trajectories.
- Real-time job market analysis and alert system for matching openings.
- Curated learning resource recommendations tied to identified skill gaps.
10. Acknowledgements
Career Mentor AI was built during HackWithBangalore 2.0, organised by HackWithIndia. We extend our gratitude to the following projects and platforms that made this possible:
Memory System: Hindsight by Vectorize (hindsight.vectorize.io)
AI / LLM: Google Gemini API (Google AI Studio)
UI Components: shadcn/ui (ui.shadcn.com)
Framework: Next.js (nextjs.org)
Live Demo: landing-page-kohl-zeta-77.vercel.app
Repository: github.com/Apekshakv/hack_with_blr_Career-Mentor-AI
Career Mentor AI - HackWithBangalore 2.0 | Built with Hindsight + Google Gemini | March 2026
Top comments (0)