Our Inspiration: The "Last 10 Percent" Problem At a hackathon, every team has had this nightmare: it's 10 minutes to the deadline, you're uploading your demo, and the congested conference Wi-Fi makes the upload bar stall at 99%.
Now imagine you're the Haas F1 team. You have a 50GB telemetry file from the car. The team in the UK needs it now for race strategy. The track-side network is just as bad as that conference Wi-Fi. A transfer that fails at 90% is a critical failure.
This is the problem that inspired Hammer Time. Standard tools like FTP or WeTransfer are passive. They just hope the network is good. Hammer Time is an active, intelligent agent that assumes the network will be bad and plans accordingly.
What We Plan to Build: A 3-Layer Logistics Agent We are designing a smart file mover built on a three-layer agentic architecture, just like a real-world logistics team.
- The Control Agent (The Driver) This is the "brainstem" of our system, written in a high-speed language like Go or Rust. It makes millisecond-level decisions. It's the driver with its hands on the wheel, constantly monitoring the network for packet loss, latency, and jitter. The moment it detects "turbulence," it will:
Dynamically switch between TCP (for speed) and UDP (for resilience).
Change the size of data chunks.
Increase or decrease the number of parallel transfer streams.
The Predictive Agent (The Navigator) This is a Machine Learning layer that learns from past transfers. It's the navigator with a memory. It will analyze the network's "profile" at the start of a transfer and say, "I've seen this unstable 4G profile before. I'll instruct the Control Agent to start with a resilient UDP-based strategy from the beginning."
The LLM Supervisor (The Logistics Manager) This is the high-level LLM agent that communicates with the user. It's the "face" of the operation.
It Reports: "Your 50GB transfer has hit congestion. The 'Driver' has adapted, and we are now proceeding in resilient mode. No data will be lost."
It Takes Orders: The user can give it commands like, "A new 10MB file was just added. That's a 'Priority One' file." The Supervisor will understand and instruct the Control Agent to pause the large transfer and get the critical file through immediately.
This multi-agent system is how we plan to build Hammer Time—a file mover that is fast, resilient, and truly intelligent.
Built With
- docekr
- fastapi
- go
- grpc
- langchain
- python
- react
- rust
- scikit-learn
- tcp/ip
- tensorflow
- udp
- websockets
Log in or sign up for Devpost to join the conversation.