PM Internship AI:
find a direction.
Practise the next step.

Discovery and preparation
belong in the same journey.

An opportunity list answers “What is available?” A candidate profile adds “What might be relevant?” Interview practice then asks “How clearly can I explain what I bring?”

The project brings those questions together. Its starting point is the candidate: education, skills, interests, languages, and location preferences. The interface then carries that context into discovery and preparation.

A useful profile,
one decision at a time.

The profile form is divided into four stages: personal information, education, skills and interests, and location. A progress indicator keeps the current stage visible, while back and next controls preserve the entered information.

Technical skills, soft skills, languages, and interests use selectable choices. Location captures the candidate’s home state and preferred destinations, making geography part of the opportunity conversation.

Keep the choices connected.

The candidate dashboard groups interests, career-path suggestions, profile improvement, and interview history. It is a place to return to after discovery and practice, rather than a separate destination with unrelated information.

The reviewed prototype includes sample career paths and match percentages. These demonstrate the interaction model; they do not establish the accuracy of an AI recommendation system.

Compare the opportunity.
See what the role asks for.

Opportunity cards combine role, organisation, location, stipend, duration, and requirements. The search component supports keyword matching across role details, plus filters for state, industry, education stream, career path, sector, work type, duration, and stipend.

Live opportunity routes read records from Supabase when configured. For the screenshots below, the original cards and interview components were rendered with the repository’s own sample records in a local capture route.

A chance to rehearse.
A clear next improvement.

The basic interview experience begins with language selection and instructions. Each question has visible progress and timing, with playback and a choice between typing an answer and recording speech.

The interface offers English, Hindi, Marathi, Bengali, and Tamil. Browser speech recognition and recording support the spoken path; typed responses remain part of the same question flow.

Feedback needs more than a number.

The grading routes request an overall score, category scores, strengths, areas for improvement, and a final assessment. Categories include communication, technical knowledge, problem solving, role fit, and confidence and clarity.

The basic route can use a response-length heuristic when Gemini is unavailable. That fallback keeps the prototype responsive, but its score is an approximation. Validating grading consistency and making the scoring source explicit are further evaluation steps.

Candidate context,
carried into the interview.

The application uses Next.js 15’s App Router, React, and TypeScript. Tailwind CSS, glass-style cards, and Framer Motion support the interface. Supabase stores candidate and internship records, with additional tables for interview sessions, questions, answers, and scores.

Generate questions from the role.

The advanced generation route reads internship details, metadata, and weighted competencies. It creates a session, asks Gemini for questions with competency information and rubrics, and stores up to five generated questions against that session.

Keep the answer and assessment linked.

The advanced grading route fetches a stored answer and its associated question and rubric. It requests structured feedback, normalises the score fields, and stores the assessment. The dashboard then joins recent sessions to their internship and score records.

Separate the demo from the live service.

Live interview routes depend on Supabase records and provider credentials. The local captures use the original components with sample inputs, so they show the experience without creating candidate accounts, submitting applications, or presenting a live AI evaluation.

An internal win.
A place at SIH ’25.

The project won IITM’s internal selection and was selected for Smart India Hackathon 2025. It presents a connected candidate experience: describing a profile, exploring roles, practising an interview, and reviewing feedback.

The next step is a measured learning loop.

The prototype establishes the interaction and integration paths. The next evaluation would test whether candidates can find relevant roles and whether repeated practice improves answer quality, with human reviewers checking feedback consistency.

The repository does not establish placement rates, recommendation accuracy, or candidate adoption. The outcome documented here is the product prototype and its IITM internal win and SIH ’25 selection.