Integrated audio flashcards into Xavier’s student app, enabling visually impaired students to create and access customized learning content.
Conducted a UX audit to identify and propose gamified learning components and interactive progress tracking dashboard.
Designed a responsive website tailored to the unique needs of mental health clients, ensuring seamless access to therapy services, resources, and appointment bookings.
Designed an interactive app with gamification and augmented reality to guide Gen Z and millennials through Mumbai’s neighborhoods, landmarks, and cultural heritage.
Built a network monitoring system for an enterprise ISP, with intuitive dashboards to visualize key metrics and predictive analytics for proactive management.
Built a speech emotion recognition system to identify of emotional tones to design a more empathetic user experience
Developed a topic modeling framework to analyze unstructured textual data and uncover recurring themes and topics in media narratives
Designed a platform to facilitate deep, meaningful conversations among the artist community, encouraging constructive dialogue and mutual support
Github Repository for this project
What if machines could hear how we feel? Speech Emotion Recognition (SER) dives into the world of vocal cues, capturing emotional undertones in spoken language. From anger to calmness, fear to happiness, this field explores how our voices convey what words alone cannot.
Emotion is at the heart of human connection, and it’s time our machines caught up. By recognizing emotions in speech, SER transforms how we interact with technology. Think virtual assistants that soothe frustration, educational tools that sense disengagement, or mental health systems that listen—not just to what you say, but how you say it.
I set out to bring emotions into the digital fold. Using the RAVDESS dataset (a treasure trove of over 7,000 emotion-labeled speech samples), I focused on identifying anger, calmness, fear, and happiness in voices.
The system demonstrated a strong ability to classify emotions, showcasing how audio feature extraction and machine learning can decode emotional states with precision. The trained model highlighted the potential of SER to drive more empathetic and adaptive systems.
From empathetic virtual assistants to smarter customer service systems, SER is already rewriting the rules of interaction. It could shape:
The journey doesn’t end here. Expanding to include diverse accents, languages, and emotional expressions will only make SER more robust. Add in continuous learning, and the possibilities grow—systems that truly understand how we feel, wherever we are.
Github Repository for this project
Ever wondered how machines sift through endless text to find meaning? Topic modeling is the answer. It’s an NLP technique that discovers the hidden themes in large collections of text. Using Latent Dirichlet Allocation (LDA), it organizes unstructured data by assigning documents to a mix of topics, each made up of related words. Think of it as sorting chaos into clarity.
The world runs on data, but raw text is messy. Topic modeling transforms that mess into actionable insights. Whether it’s organizing reports, uncovering trends, or simplifying information retrieval, LDA gives structure to the unstructured, making it indispensable for tasks like media analysis, corporate strategy, or policy review.
With over 500 H2020 reports in hand, I set out to uncover the core topics they addressed. By combining advanced text processing with topic modeling techniques, I built a framework to map out the diverse areas covered by these initiatives.
The result? A detailed map of the key themes embedded within the reports. From topic relationships to the most prevalent terms, the framework delivered an intuitive way to navigate the data and extract meaningful insights.
From empathetic virtual assistants to smarter customer service systems, SER is already rewriting the rules of interaction. It could shape:
This is just the start. By incorporating domain expertise, the model could better interpret specific contexts. And with real-time capabilities, it could process new documents as they arrive, keeping insights fresh and actionable.
What happens when artists step away from the noise and come together for genuine dialogue? DeepTalks is an Android application designed to spark these conversations. Inspired by “Guftagu Karein?”, an Indian card game for close-knit discussions, DeepTalks takes this concept into the digital world, creating a space where artists can reflect, share, and connect.
It’s simple. One person creates a session by generating a game code. Others join using the same code, unlocking a shared experience. Everyone sees the same image and prompt simultaneously, setting the stage for unified, thoughtful conversations.
At its core, DeepTalks thrives on carefully crafted prompts. Think: “The price of passion,” “Navigating rejection,” or “Defining moments.” These aren’t just questions—they’re invitations to explore untold stories, personal insights, and collective experiences.
In a world of shallow social media interactions, DeepTalks offers something different: depth. It turns ordinary gatherings into hubs of connection, helping artists articulate their emotions, challenge assumptions, and foster a sense of belonging. It’s not just about talking—it’s about being heard.
DeepTalks transforms creative isolation into community. By providing a platform for reflection and shared experiences, it encourages vulnerability, empathy, and collaboration. For artists, it’s more than a game—it’s a reminder that connection is as vital as creativity.
DeepTalks is just getting started. Imagine customizable prompts, AI-generated conversation starters, or features tailored to specific creative communities. The possibilities are endless—but the goal remains the same: meaningful connection in a disconnected world.
