Watson

🧠 Inspiration

In the fast-paced world of networking, meetings, and spontaneous conversations, important insights often get lost in the noise. Whether it's a meaningful chat at a conference, a brainstorming session at work, or a chance encounter with someone who could be a valuable connection, we rarely capture the full context of what was said—or who said it.

We were inspired by the idea of augmenting human memory using AI—specifically, creating a tool that doesn't just record conversations, but actually helps users remember, reflect, and reconnect. Imagine finishing a conversation and having a clean, clear summary of what was discussed, along with the names and LinkedIn profiles of the people you spoke with—all in one place. That’s what Watson aims to provide: a personal AI-powered conversational memory companion for the real world.

✅ What it does

Watson is a mobile app that quietly listens to your conversations, and once the discussion ends, it goes to work:

Transcript Summarization: Watson distills long conversations into concise summaries, capturing key takeaways, decisions made, action items, and any notable moments.
Speaker Identification: It analyzes the conversation to identify distinct speakers and uses AI to suggest their likely LinkedIn profiles, so you can follow up or build your network.
Post-Conversation Briefing: After each interaction, users receive a summary card that includes:
- A quick bullet-point summary
- A list of speakers
- Suggested LinkedIn profiles (where applicable)
- Tags or topics extracted from the conversation
Conversation Archive: Users can view a timeline of past conversations with full transcripts, summaries, and speaker details—like a searchable, smart memory vault.

💻 How we built it

Frontend (Mobile App): Built with Expo and React Native to ensure smooth cross-platform performance and a lightweight user interface optimized for quick access post-conversation.
Backend & AI Orchestration:
- Powered by Gemini's agentic flow, which handles:
- Transcript parsing and summarization
- Entity recognition for speaker names
- LinkedIn profile suggestion and verification
- Audio is transcribed (via speech-to-text) and then parsed by Gemini to generate conversational insights in near real-time.
Database:
- MongoDB Atlas stores:
- Raw and processed transcripts
- Structured speaker profiles and their metadata
- Summaries and historical conversation context

🤔 Challenges we ran into

Voice Identification: Accurately identifying speakers was a core challenge. We avoided biometric identification, opting instead for conversational cues and names mentioned aloud.
LinkedIn Matching Accuracy: Finding the right LinkedIn profiles based on name, job title, or topic context proved technically nuanced, especially for common names. We had to fine-tune our Gemini prompts for precision and reliability.

🎉 Accomplishments that we're proud of

We successfully created an AI-powered companion that transforms everyday conversations into actionable insights and memory anchors.
The seamless integration of Gemini for multi-step agentic tasks, especially speaker identification + LinkedIn matching, was a major technical win.

🎓 What we learned

Building an agentic AI system means carefully designing flows that feel intelligent, but also predictable and transparent. Prompt chaining and output parsing were surprisingly tricky.
Context matters: The more surrounding detail we gave Gemini (like meeting context or known names), the more accurate and relevant its outputs became.
We gained hands-on experience balancing mobile UX with AI backend complexity, and saw firsthand how crucial it is to surface insights without overwhelming the user.

🚀 What's next for Watson

Smart Contact Sync: Automatically build a contact book with contextual notes from each conversation.
Team Mode: Enable teams to sync shared meetings and post-summary insights into tools like Slack, Notion, or Google Docs.
In-App LinkedIn Messaging: Allow users to reach out directly from the summary card once a LinkedIn profile is confirmed.
Offline Mode + Local AI: For users concerned about privacy, we plan to experiment with on-device processing using edge LLMs and Whisper.

Ultimately, we envision Watson as your AI memory, helping you stay thoughtful, prepared, and connected—one conversation at a time.