ERA

Master the Voice. Secure the Grant. Win the Future. -The ultimate Gemini-powered partner for international success.

Comment

Inspiration

💡: From Conflict Zones to Campus Master the Voice. Secure the Grant. Win the Future. -The ultimate Gemini-powered partner for international success. This project is not just code to us; it is our lived reality. We are Sakib and Rawa, two international students hailing from Bangladesh and Yemen. Coming from regions scarred by war, political instability, and economic fragility, we learned a harsh truth early on: Education is the only exit strategy.

For students like us, obtaining a Higher Study degree abroad isn't a luxury—it's a lifeline. However, once we arrived, we hit two massive, invisible walls that threatened to derail our future: The Fluency Trap: High-Stakes Speaking Anxiety (The Silence): We knew the English grammar and vocabulary, but we froze in high-stakes environments. Whether it was the pressure of an IELTS speaking examination, a critical visa interview, or even a professional networking opportunity, we lacked a realistic, human-like practice partner. Static language apps don't interrupt your hesitation, critique your tone, or simulate the pressure of a real human examiner who holds your future in their hands. The problem: Millions of talented students are silenced not by a lack of knowledge, but by the anxiety of speaking under pressure, with no accessible tool to truly simulate the test environment.

The Funding Maze: Scholarship Fragmentation and Verification Crisis (The Maze): Securing verified, relevant scholarships was our only way to fund our degrees and survive abroad. Yet, we were forced to navigate a labyrinth of thousands of outdated, unstructured websites and portals. Finding verified opportunities for students from the Global South—especially those with limited and inconsistent internet access—was a full-time job. We spent more time chasing broken links and vetting scam-riddled lists than applying for the grants that could secure our future. The problem: Critical, life-changing scholarship data is fragmented and unverified, creating a massive barrier for the students who need it most. We realized millions of brilliant minds in conflict zones and developing nations are silenced not by a lack of talent, but by a lack of the right tools and verified information. ERA was born to break that silence. We wanted to build the mentor we never had: one that listens, corrects, and guides us to the funding we deserve.

What it does

ERA is the world's first Adaptive Academic Ecosystem powered by Google Gemini 3, providing a single, custom-configured, and adaptive environment that uniquely bridges the gap between language mastery and academic opportunity. Its comprehensive capabilities are delivered through two core engines and an overarching adaptive intelligence: The Vocal Mastery Hub (IELTS Simulator & Tailored Assistant): IELTS Simulation: Unlike static language apps, ERA uses Gemini's Multimodal Live capabilities to function as a truly dynamic and personalized vocal coach. It can instantly embody a strict IELTS examiner for high-stakes practice, offering instant feedback on Fluency, Lexical Resource, and Pronunciation. Tailored/Casual Voice Assistant: The same core technology allows ERA to shift roles instantly to act as whatever the user needs—be it a friendly conversation partner, an HR interviewer, a critical feedback provider, or a supportive mentor. This provides a holistic, non-IELTS-specific environment for general language practice, casual chat, and customized communication scenarios, eliminating the need to switch apps for varied practice. The Scholarship Scout Engine: Intelligent Scholarship Search: This is a neural search engine designed to connect a student's profile (GPA, Country of Origin, Major) to a massive database of global funding opportunities, cutting through the noise to find specific grants, especially for students from underrepresented regions. Voice-Guided Information & Assistance: Crucially, this engine integrates with the voice capabilities. Users can speak to the "Scout" to ask detailed questions about the funding, the country, the application process, and any other aspect of the opportunity they've selected. It provides spoken, detailed explanations and guidance, acting as an interactive advisor, not just a search tool. In summary, ERA is a versatile ecosystem that is not only for IELTS and scholarship searching, but also for: Tailored & Casual Chat: Providing a dynamic, conversational AI assistant for general communication practice and personalized support. Voice-Guided Information: Delivering detailed, on-demand spoken information and assistance regarding scholarship opportunities, countries, and application requirements.

How we built it

We leveraged the cutting-edge capabilities of Google Gemini 3 to power the intelligence of ERA, demonstrating the model's immense power and versatility. The entire system was developed rapidly using Google AI Studio and a collaborative, high-energy approach we call "vibe coding." Gemini 3 Power: The core intelligence of ERA is driven by the powerful Gemini 3 model, which provides sophisticated reasoning and complex information processing capabilities. Development Environment: The rapid prototyping and configuration of ERA were managed efficiently within Google AI Studio, allowing us to seamlessly integrate and deploy Gemini's advanced features. Multimodal Live API (The Voice - Adaptive and Real-time): We bypassed traditional "Speech-to-Text $\rightarrow$ LLM $\rightarrow$ Text-to-Speech" pipelines, which are often too slow for natural conversation. Instead, we stream audio data directly to Gemini, enabling ERA to understand how a user is speaking (nervousness, pauses) and not just what they are saying, leading to a much more natural and empathetic interaction. Live Information Retrieval: Our system configuration is highly adaptive. The scholarship search function (sout) leverages the Gemini Multimodal Live API to fetch up-to-the-minute, valid scholarship information and direct application links, ensuring students receive the most accurate and current resources. Tech Stack: React, WebSockets, Gemini Multimodal Live API. Long-Context Window (The Brain - Massive Data Ingestion): Scholarship data is inherently messy—consisting of thousands of PDFs, unstructured university bulletins, and government criteria documents. We utilized Gemini’s massive 1M+ token context window to ingest this vast amount of unstructured scholarship data simultaneously, providing a comprehensive and deep understanding of the criteria. Algorithm: We designed a precise matching function where $S$ (Student Profile) interacts with $C$ (Criteria Corpus) to maximize the Probability of Acceptance ($P_{acc}$): $$P_{acc} = \text{GeminiReasoning}(S, C_{\text{unstructured}})$$ System Instructions (The Persona - Dynamic Empathy): We employed advanced prompt engineering to create "Adaptive Personas." The system dynamically shifts its tone and communication style based on the user's stress levels and conversational context, ensuring the simulation feels genuinely human, empathetic, and supportive throughout the process.

Challenges we ran into

  1. Implementing "Vibe Coding" into Reality (The Latency Battle): Initial Problem: Early versions of the real-time conversational tool suffered from a disruptive 3-4 second delay, which broke the sense of natural dialogue. The Concept of "Vibe Coding": This term was coined to define the goal: an interaction flow so smooth and responsive that it feels like a genuine, sub-second, human conversation. Solution - Technical Optimization: Optimized WebSocket handling for more efficient communication. Fine-tuned Gemini's streaming settings not just for raw speed, but specifically for the rhythm required to achieve near-human conversational speed. Outcome: Achieved a flow that felt natural and immersive, realizing the "Vibe Coding" goal.
  2. Proper Prompting for Grounding (Hallucination in Funding): Initial Problem: The AI model (Scholarship Scout) showed a tendency to "hallucinate"—inventing or misreporting critical facts, especially details like application dates or deadlines, which is high-stakes for users. The Solution - Robust Grounding Step: This was achieved by implementing an extremely specific and proper prompting strategy. Key Instruction in the System Prompt: The core prompt now strictly and explicitly instructs the Gemini model to: Cite Sources: Reference the provided context for all facts. Verify Dates: Cross-reference and confirm every date against the source material. Outcome: Reduced hallucination to near zero, making the information provided by the AI trustworthy and reliable.
  3. Proper Configuration for Adaptive Scenarios and Flexibility with User Custom Prompts (Emotional Weight): Initial Problem: The tool needed to serve students from diverse, often war-torn regions, requiring sensitivity beyond a typical, robotic AI tone. The Solution - Flexible Emotional Configuration: A system called the "Empathetic Mentor" was designed to adapt its tone and response style. Key Implementation Details: Multiple Adaptive Scenarios: The system was configured with various pre-defined scenarios that account for different levels of emotional weight (e.g., standard inquiry vs. expressions of distress). Corresponding Prompt Variations: Each scenario triggers a specific variation of the system prompt to guide the AI's response, ensuring it is empathetic and personalized, not dismissive. Flexibility with User Custom Prompts: The system's design specifically accommodates complex or deeply distressing input from the user (custom prompts) by ensuring the tone automatically flexes to match the detected emotional context. Outcome: Ensured the AI can navigate highly sensitive topics with an appropriate, empathetic tone, providing a personalized and supportive interaction.

Accomplishments that we're proud of

Global Impact and Pride: We are immensely proud that we have successfully developed a powerful tool designed to assist students all over the world, extending our reach and support globally. Invaluable Knowledge Gained: Throughout the development process, we have acquired invaluable knowledge and expertise, significantly deepening our understanding of the challenges and solutions in educational technology. Comprehensive Problem Resolution: We successfully addressed and fixed all previously mentioned problems and challenges, ensuring a robust and reliable final product. Creation of a Custom AI Assistant: A major achievement is the creation of a truly adaptive and custom AI assistant, specifically tailored to meet the unique learning needs of the future generation of dreamers. Vision Realized and Smooth Operation: We are deeply satisfied and proud that our initial vision has come to fruition. Everything is currently working smoothly and effectively, marking the successful culmination of our efforts.

What we learned

Mastering Multimodal Interaction (Including Current Point): We confirmed that the future of complex AI applications, particularly in domains like advanced education, necessitates moving beyond text-only models. We learned that the true essence of communication—the nuances of tone, pitch, and speed—are crucial data points for a holistic understanding. This directly underscored the essential capability of Gemini's multimodal architecture to process diverse inputs, offering a significantly richer and more human-like interaction than simpler models. The Unprecedented Power of Long-Context Understanding (Including Current Point): A core takeaway was the discovery of Gemini's industry-leading long-context capabilities. By successfully feeding the model extensive and highly detailed documents (like entire university handbooks), we demonstrated that its architecture does not falter under the weight of vast information. Instead, it uses this depth to significantly enhance accuracy and perform deep, reliable information retrieval—a key differentiator from previous generation models. The Art of "Vibe" Coding and Intent Capture (Learned Vibe Coding): We gained critical experience in what we termed "vibe" coding. This is the subtle but vital skill of iterating and fine-tuning prompts not just for literal instructions, but to successfully capture the intent, desired persona, and feel of a complex task. This mastery of prompt engineering, focusing on the underlying "vibe," is essential for unlocking the most advanced and nuanced responses from large, sophisticated models. High-Pressure, Rapid-Iteration Teamwork (Learned Teamwork): The hackathon environment provided invaluable, hands-on experience in high-pressure, rapid-iteration teamwork. We learned how to quickly transition from an abstract concept to a functional prototype, efficiently leveraging individual and diverse team strengths. This accelerated process honed our ability to collaborate effectively under tight deadlines and technical constraints. Bridging the AI Expectation-Capability Gap (The Gap Between AI and Student): We gained direct insights into the current disconnect between student expectations of AI functionality and the actual, practical capabilities and limitations of existing models. This understanding is foundational for future development, ensuring that our AI applications are designed to be practical, user-centric, and truly address real-world student needs rather than chasing unrealistic, sci-fi expectations. Adaptive and Resourceful Problem-Solving (Adaptiveness): The constant, dynamic nature of working with cutting-edge technology required us to master adaptive problem-solving. The need to rapidly pivot our approach in response to new model insights, unexpected technical hurdles, or evolving scope honed our skills in flexible and resourceful engineering within a highly advanced technological domain. Future Capabilities and Impact of Advanced Models (Capabilities of Gemini 3 & What It Can Do for Future): Our work highlighted that models like Gemini represent a fundamental shift in AI capability. Their long-context and multimodal power enables future applications that can: Serve as a true, comprehensive digital expert: Capable of synthesizing entire bodies of knowledge (e.g., all university curriculum) into digestible, personalized learning pathways. Understand complex human communication: Utilizing multimodal input to teach language and communication with a depth previously impossible. Facilitate hyper-personalized education: Dynamically adapting to a student's individual learning style, pace, and current emotional state (via tone/vibe analysis) to provide the most effective intervention.

What's next for ERA

The next phase of ERA's development is driven by a deep commitment to maximizing our impact. We are incredibly passionate about this project and firmly believe that with the necessary support, we can launch a full-scale application that will profoundly benefit millions of students worldwide. Our immediate focus includes: Video Interview Analysis powered by Gemini: We plan to leverage Gemini's cutting-edge advanced vision capabilities to analyze a user's non-verbal communication—specifically eye contact, facial expressions, and body language—during mock interviews. This sophisticated analysis will provide highly personalized, actionable feedback, dramatically helping students refine their presentation skills and boost their confidence for crucial real-world job and university interviews. Direct University API Integration for "One-Click Apply": We aim to establish direct partnerships with universities to integrate our platform with their application systems. This powerful integration will enable students to utilize a "One-Click Apply" function directly from the ERA dashboard. The system will automatically populate and submit their applications using the verified information already stored in our platform, creating an unparalleled, highly streamlined, and efficient application process. Offline Mode for Universal Access: Recognizing the global disparities in connectivity, especially in conflict zones or remote areas, we are developing a lightweight, locally stored version of key resources and essential career development tools. This crucial 'Offline Mode' ensures continuous, uninterrupted access to educational materials, regardless of unstable or non-existent internet connections, upholding our mission of universal accessibility. Document Auto-Drafting: Using the user's profile to automatically generate the first draft of their "Statement of Purpose" or "Motivation Letter" tailored to the specific scholarship found. University Partnerships: Allowing universities to host their scholarship data directly on ERA for verified, real-time access.

Built With

Share this project:

Updates