Jarvis - Autonomous Context & Workspace Copilot

Inspiration

Modern professionals waste hours context-switching between fragmented workspaces, buried emails, and siloed notes. We envisioned Jarvis: a private, autonomous personal chief of staff that seamlessly bridges static knowledge and live communications into a single intelligent interface.

What it does

Dynamic Workspace RAG: Indexes local project files, notes, and technical documentation into high-dimensional vector representations using Gemini embeddings.
Live Communications Search: Connects securely to communication channels (like Gmail via App Passwords) to fetch contextual email threads on demand.
Agentic Tool Orchestration: Leverages the Google Agent Development Kit (ADK) and Gemini models to autonomously route queries, inspect evidence, and synthesize factual answers.
Secure & Cloud-Ready Logging: Deployed with Google Cloud Run.

How we built it

AI & Agent Core: Built using Google ADK and the official Google GenAI SDK, running inference on gemini-2.5-flash / gemini-embedding-2.
Frontend UI: Crafted a responsive, zero-credential-leakage interface using Streamlit.
Cloud Infrastructure: Containerized with Docker and hosted serverless on Google Cloud Run, backed by Google Cloud Firestore for session logging.

Challenges we faced

Privacy & Secret Isolation: Designing a stateless architecture where sensitive API keys and personal credentials remain purely in the user's browser session without persistent disk exposure.
Hybrid Context Fusion: Balancing static indexed knowledge chunks alongside real-time live API retrieval while minimizing latency.