I am a student and also a competitive exam aspirant, having appeared for a few exams, I understand the weightage current affairs carries in these exams. I am fond of reading newspapers, but when you are preparing for an exam reading and watching news to figure out potential questions can be very overwhelming. One morning I was reading news from various sites like reuters, the hindu, new york times and that's when it struck me, I thought of using AI to provide me with news which was targetted as potential current affairs question and that's when I decided to create CurrentIQ.
CurrentIQ is an AI-powered current affairs engine which helps aspirants to prepare for their exams in an efficient and time-saving manner.
Backend: We built a Python FastAPI server that exposes three main endpoints — /digest, /mcq, and /evaluate. The server aggregates live news from multiple RSS feeds including PIB (Press Information Bureau), The Hindu, BBC India, DD News, and Economic Times using the feedparser library. Articles are deduplicated using a fingerprinting approach before being sent to the AI.
Frontend: Built with React and Vite. Three main pages — DigestPage, MCQPage and EvaluatorPage — all communicate with the FastAPI backend via axios. The UI features a Today's Radar panel showing live headlines, exam selector buttons, and a clean dark theme.
- RSS feed reliability: Some RSS feeds like ISRO were slow or blocked from certain networks. We solved this by adding multiple fallback feeds and tagging each feed with exam relevance so each exam always has enough articles even if one feed fails.
- News deduplication: Multiple sources often report the same story with slightly different headlines. Simple string matching over or under-merged stories. We built a word fingerprinting approach that uses the first six significant words of each headline as a unique identifier, which gave much better deduplication results.
- Prompt engineering for six different exam formats: Getting Nova to produce structurally different outputs for six exams required careful prompt design. Each exam prompt had to be specific enough to force the right format but flexible enough to handle any news topic. This required multiple iterations to get right.
Getting Nova to return consistent JSON: Nova sometimes returned JSON wrapped in markdown code fences or with preamble text before the opening bracket. We built a robust parse_nova_json() function that strips markdown fences using regex, attempts direct JSON parsing, and falls back to finding the first valid JSON object in the response.
Exam-aware AI repackaging: The fact that the same live news article gets transformed into completely different formats for different exams is our biggest technical accomplishment. A story about India's space program becomes an analytical GS3 note for UPSC, a rapid-recall fact sheet for NDA, and a technical summary for GATE — all from a single Nova API call with different prompts. This works reliably across all six exams.
UPSC Answer Evaluator: Building an AI evaluator that scores answers on six official UPSC Mains criteria and returns structured, actionable feedback is something no free tool currently offers. Seeing Nova reason about the quality of a written answer — identifying missed points, suggesting improvements, and writing a model introduction — was a genuine wow moment during development.
Full agentic pipeline: The platform is not just a chatbot wrapper. It autonomously fetches live news, deduplicates articles across sources, selects the most relevant stories, structures them for the target exam, and generates practice material — all in a single user action. This end-to-end agentic flow powered by Amazon Nova is something we are genuinely proud of.
Real-time news integration: The Today's Radar panel shows live headlines from PIB, The Hindu, BBC India and Economic Times updated in real time. Using PIB — India's official government press bureau — as a primary source means the content is directly relevant to what appears in actual exam questions.
Solving a real problem for a massive audience: India has 1.7 million competitive exam aspirants who spend hours daily on newspaper reading. Building something that genuinely saves them time and improves their preparation quality — and seeing it work live — is our proudest accomplishment.
Amazon Nova's reasoning capability goes beyond summarisation:
We initially expected Nova to just summarise news articles. What surprised us was its ability to reason about exam relevance — correctly identifying which GS paper a story belongs to, what constitutional provisions are linked, and how a current event connects to static GK topics. This level of domain reasoning from a general purpose model was genuinely impressive.
Prompt engineering is the core skill for agentic AI: The quality of our output was almost entirely determined by how well we wrote our prompts. Small changes — adding "plausible distractors" to the MCQ prompt or specifying "forward-looking conclusion" in the evaluator prompt — had dramatic effects on output quality. We learned that with a powerful model like Nova, prompt design is the most valuable skill.
Async architecture matters for AI applications: Wrapping synchronous boto3 calls in asyncio executors taught us a lot about how to build production-ready AI backends that stay responsive under load. This pattern is essential for any real-world AI application built on AWS.
Public domain sources are underrated: We initially planned to scrape paywalled newspapers. Switching to PIB and government RSS feeds not only avoided legal issues but actually improved content quality for exam purposes — government press releases are directly more relevant to UPSC than newspaper opinion columns.
The value of domain-specific AI: Building for a specific audience — Indian competitive exam students — rather than a general use case made every feature more focused and more useful. Amazon Nova gave us the intelligence, but understanding the UPSC exam pattern deeply is what made the product genuinely valuable.
Speed of building with AWS Bedrock: We were able to go from idea to working prototype extremely quickly using Amazon Bedrock. Not having to manage model infrastructure and getting immediate access to Nova 2 Lite let us focus entirely on building the product rather than managing AI infrastructure.
What's next for CurrentIQ -
Voice Digest using Amazon Nova 2 Sonic: The next immediate feature is a voice-based current affairs briefing using Amazon Nova 2 Sonic. Students could listen to their personalised exam digest during their commute, turning dead time into study time. Nova 2 Sonic's natural conversational voice would make this feel like a personal tutor speaking to you rather than a text-to-speech robot.
Personalised study plans: Track which topics a student is weak in based on their MCQ performance and answer evaluations, then use Nova to generate a personalised 30-day current affairs study plan targeting their weak areas.
Previous Year Question analysis: Feed UPSC and NDA previous year question papers to Nova and identify patterns — which topics appear most frequently, which current events are likely to be asked this year. Give students a predicted hot topics list before their exam.
Multi-language support: Translate digests and MCQs into Hindi and other regional languages using Nova's multilingual capabilities. A large portion of UPSC aspirants prefer Hindi medium and currently have very limited access to quality AI tools in their language.
Mobile application: Build a React Native mobile app so students can access their daily digest, attempt MCQs, and submit answers for evaluation on their phone. Push notifications for breaking news relevant to their exam.
Classroom integration: Partner with offline coaching institutes across India to integrate CurrentIQ into their daily teaching workflow. Teachers could use the digest as a starting point for current affairs classes and assign MCQs as daily homework.
Expanding beyond India: The same architecture works for any country with competitive examinations — UPSC equivalent exams exist in the UK (Civil Service), USA (Foreign Service), and across Southeast Asia. CurrentIQ could become a global platform for government exam preparation powered by Amazon Nova.
Log in or sign up for Devpost to join the conversation.