AI Admission Counselor.
An AI-powered admission counseling platform that automates personalized engineering-stream recommendations — processing student academic data, generating adaptive assessment questions, and delivering stream recommendations as polished PDF reports.
How It Works
- Students complete a dynamic interest questionnaire where questions adapt based on prior answers; responses combine with academic scores.
- A FastAPI backend runs a RAG pipeline with a prompt-engineered GPT-4 layer to generate personalized questions and counseling reports.
- Responses stream to the client for a real-time report-generation experience.
- An automated Markdown → PDF pipeline turns the generated report into a polished, shareable document.
Architecture & Key Decisions
- RESTful API with JWT (RSA256) authentication, versioned endpoints (v1), and role-based access.
- Retrieval-augmented generation grounds the LLM in each student's academic profile and questionnaire responses before generating recommendations.
- Secure file-upload system with automatic WebP conversion and MIME validation.
- Admin dashboard for lead management, counseling-session tracking, and reporting.
Why It Matters
Production LLM integration in a real client product
This wasn't a demo — it was a shipped product replacing a manual counseling workflow. It required real prompt engineering, a retrieval pipeline to keep recommendations grounded in each student's data, streaming responses for UX, and structured report generation reliable enough to hand to families making a decision.
Trade-offs & What v2 Adds
Why synchronous OpenAI calls?
Simpler implementation with acceptable latency for the use case.
Why file-based key storage?
Acceptable for the MVP versus standing up a full KMS.
What v2 would add
Redis caching, an async queue (RabbitMQ/SQS) for OpenAI calls, S3 for uploads, and rate limiting.
Want the full walkthrough?
Happy to walk through the RAG pipeline, prompt design, and streaming report generation in detail.