# Career Jyoti × Job Mela ## Project Brief Career Jyoti and the Job Mela platform take a student from career confusion to a real job opportunity in one continuous experience. Career Jyoti is an AI-powered career guidance chatbot built for students without a strong career or technical background — including students from the education sector — who can ask about career planning, market trends, or next steps and get a clear, direct answer instead of generic advice. Built directly into that same chat experience is the Job Mela system: a platform for running job fairs across India, where students browse participating companies, register, apply to multiple roles, and track their status, while organizers run the entire fair — from onboarding companies to final shortlisting — through a dedicated admin app. I designed and built both systems end-to-end — database, backend, mobile, and web — working directly with the client throughout. Career Jyoti was built with one teammate; the Job Mela platform, which has since run two live fairs and onboarded 6,000+ students, was built solo in about two months, from the first architecture decision to production deployment. ## Problem Statement Students without a strong technical or career-focused background often don't know where to turn for reliable career advice. Questions about which roles suit them, what the job market looks like right now, or how to plan the next few years usually go unanswered — or get generic, one-size-fits-all responses that don't reflect their actual situation. Human career counseling simply can't scale to answer this for every student who needs it, instantly and individually. Even when students do get good advice, there's usually a gap between guidance and opportunity. Career direction and real hiring openings tend to live in two completely separate places, with no direct path from one to the other. Running a job fair at scale is a separate, equally hard problem. Coordinating dozens of companies, thousands of applicants, multiple roles and interview rounds per company, and tracking who's shortlisted or selected — all while the fair is actively happening — is normally a heavy, manual, error-prone process. And at real scale, issues that never surface in testing show up immediately, with no room to fail — exactly what happened once registrations crossed 6,000 for a single event. ## Solution Architecture The two systems work as a single, connected journey rather than two separate tools. A student opens Career Jyoti and asks a normal career question; behind the scenes, the bot first checks a curated bank of roughly 2,000–3,000 pre-written Q&A pairs for a fast, dependable answer, and only falls back to an AI-generated response when nothing in the bank matches closely enough — so the student never hits a dead end. The bot also tracks what was asked earlier in the conversation, so follow-up questions build on what came before instead of starting from zero. Whenever a job mela is live, a registration option appears directly inside the chat. A student can move straight from career guidance into registering for the fair, then choose which companies and roles to apply to, up to a set maximum per student so the process stays fair under high demand. Everything is visible in real time to the organizing team: every registration, application, and status change shows up instantly in the admin app, and once a student is marked selected or shortlisted, that outcome is locked in and their remaining open applications adjust automatically. On the organizing side, each job mela gets its own secure, private sign-up link that the organizing team uses to build out the list of participating companies, roles, and interview rounds before the fair opens to students. Once live, an analytics dashboard gives organizers and leadership visibility into a single fair or all fairs combined — company participation, applications, and selection progress, down to the level of an individual role. ## Technical Execution **Career Jyoti** - Built the response engine as a two-layer system: a fast lookup against a curated bank of ~2,000–3,000 Q&A pairs, with an AI model stepping in only when nothing in the bank matches closely enough - Built the conversation-context pipeline so follow-up questions build on what was asked earlier in the chat - Trimmed how much text is sent to the AI model per call, keeping response costs down without hurting answer quality - Validated the system internally with 30+ testers across the company before launch, including fallback-generated answers **Job Mela** - Owned every layer solo — database design, backend logic, load handling, the admin mobile app, and the web experience — built from scratch in about two months - Engineered the backend to handle high-volume, fast-changing registration data, proven to handle 500+ requests per second - Secured the admin app with Google-based login; built student registration around a mobile number and email, with sessions active for about 18 hours - Logged every admin action — present, absent, shortlisted, selected — for a full activity trail per fair - Built real-time seat and capacity tracking, with capacity guarded from ever being set below the number of students already registered - Added a background reconciliation process to catch and correct registration-count drift after bulk imports or cancellations - Diagnosed a mobile app slowdown that surfaced only once thousands of real students hit the platform at the Lucknow Job Mela (August 2026), and shipped a caching layer that resolved it live ## Results & Impact - **2 job melas** run successfully on the platform, with the **Lucknow Job Mela (August 2026)** as the first live trial - **6,000+ students** registered and actively used the system - Platform proven to handle **500+ requests per second** under real usage - Career Jyoti's responses — including AI fallback answers — validated with **30+ internal testers** before launch, with consistently positive feedback - A post-launch latency issue caused by real-world concurrent load was diagnosed and resolved with a caching layer, keeping the app usable during the event itself rather than only after it Beyond the numbers, the bigger shift is from two disconnected experiences to one continuous path: a student can go from not knowing what career is right for them to applying to companies at a live job fair, without ever leaving the conversation — while the organizing side moves off spreadsheets and manual shortlisting onto a system that keeps seats, applications, and outcomes accurate in real time. ## Key Learnings - Real usage finds problems testing can't: the mobile app's latency issue only appeared once thousands of real students were on it simultaneously during the actual event — reinforcing the value of designing with load headroom in mind and being able to diagnose and ship a fix fast, under real pressure - Answering at scale doesn't require sending every question to a large AI model: checking a curated, pre-vetted answer bank first and falling back to AI generation only when needed kept responses fast and cost-efficient without limiting coverage - Shared, limited resources like interview seats need more than a simple decrement-on-apply rule to stay trustworthy at scale: a background reconciliation step to catch drift mattered as much as the real-time tracking logic itself - Owning a project end-to-end and working directly with the client throughout made it possible to catch issues in the actual user flow — like how seat-locking should behave once a student is selected — and correct them fast, without routing decisions through several layers of people