Skip to content

PITCH_SCRIPT.md

Latest commit

 

History

History
161 lines (119 loc) · 6.13 KB

PITCH_SCRIPT.md

File metadata and controls

161 lines (119 loc) · 6.13 KB

🦅 PANOPTES: Pitch Script

Duration: 3-5 minutes
Audience: SparkHacks Judges / VCs / Public Safety Stakeholders

🎬 OPENING HOOK (30 seconds)

[Show Landing Page on screen]

"In San Francisco, it takes an average of 17 minutes for police to respond to a 911 call. In that time, situations escalate. People get hurt. Evidence disappears.

What if police were already in position before the crime even happened?

This is Panoptes — named after the all-seeing giant of Greek mythology — an AI-powered predictive policing platform that doesn't just react to crime. It anticipates it."

💡 THE PROBLEM (45 seconds)

[Stay on Landing Page or show stats]

"Today's emergency response is fundamentally reactive. Officers sit in precincts waiting for dispatch. Data exists — historical crime patterns, 311 complaints, weather, traffic, even neighborhood stressors — but it's locked in silos.

Meanwhile, cities like San Francisco spend $700 million annually on policing with declining public trust and increasing response times.

The data to predict crime already exists. We just haven't connected it yet."

🚀 THE SOLUTION (60 seconds)

[Navigate to Command Center - /command]

"Panoptes is the tactical operating system for modern policing.

[Point to the heatmap]

This is our real-time Predictive Risk Map. Every neighborhood in San Francisco is scored by our ML model — trained on 3 years of incident data, combined with 30+ real-time stressor signals including weather, traffic congestion, 311 complaints, and time patterns.

Red zones are where incidents are most likely to occur in the next hour. Officers can be pre-positioned before the first 911 call comes in.

[Click on the Comms tab to show the AI agent]

And this is AEGIS — our agentic AI interface. Officers can ask natural language questions like:

'What's the risk level in the Tenderloin right now?'
'Show me vehicle theft patterns for the weekend'
'Why is Mission District flagged as high-risk tonight?'

AEGIS queries our database, runs analysis, and returns actionable intelligence in seconds — all powered by NVIDIA Nemotron running 100% locally."

🎥 LIVE VISION DEMO (45 seconds)

[Click on Fleet tab → View Live Feed button]

"But prediction is only half the story. When officers are on scene, they need real-time situational awareness.

[Show Live Feed Modal with WebRTC video]

This is SENTINEL — our live vision system. It connects to body cameras, vehicle cams, or drones via WebRTC and runs real-time Vision Language Model analysis.

[Point to caption panel]

The AI watches the feed and generates live captions: identifying suspects, detecting weapons, counting individuals, assessing crowd behavior. When the stream ends, it generates a full incident summary — automatically.

This isn't surveillance. This is officer safety and accountability."

🔧 THE TECH STACK (45 seconds)

[Show Landing Page Architecture section or speak over Command Center]

"What makes Panoptes unique is our edge-first architecture:

  • Dell DG10 with NVIDIA GPU and ARM architecture
  • 100% local processing — zero cloud dependencies
  • Sub-10ms inference for real-time predictions
  • NVIDIA Nemotron for LLM-powered queries
  • Live VLM WebUI for real-time video analysis
  • LightGBM for risk prediction with 30+ feature signals

All sensitive data stays within the department's secure perimeter. No citizen data ever leaves the building. Complete data sovereignty.

This isn't just a prototype — it's running live on a DGX Spark right now."

📊 SIMULATION DEMO (30 seconds - Optional)

[Navigate to /simulation or show Simulation Control in sidebar]

"For training and planning, we built a Simulation Engine that can replay historical scenarios or generate future forecasts.

[Play a few frames of simulation]

Watch how risk levels shift across neighborhoods over a 10-day period as conditions change — weather, events, time of day. This lets command staff practice resource allocation before high-risk periods like New Year's Eve or major events."

📈 THE MARKET (30 seconds)

"The predictive policing market is projected to reach $15 billion by 2030. But current solutions are cloud-dependent, slow, and raise serious privacy concerns.

Panoptes solves all three:

  • ✅ Local processing — no cloud
  • ✅ Real-time — sub-second latency
  • ✅ Privacy-first — data never leaves premises

Our initial target: San Francisco PD as a pilot, then expand to Oakland, San Jose, and the broader California market."

🎯 THE ASK (20 seconds)

"We built this in 48 hours at SparkHacks. With additional funding, we can:

  1. Complete the cuOpt integration for optimal unit routing
  2. Expand training data to include real-time social signals
  3. Launch a 6-month pilot with SFPD Tenderloin Station

We're looking for partners who believe public safety deserves the same AI innovation as every other industry."

🔥 CLOSING (15 seconds)

[Return to Command Center with heatmap visible]

"Panoptes doesn't replace officers. It makes them smarter, faster, and safer.

This is the future of public safety. And it's running — right now — on a DGX Spark.

Thank you."

📱 QUICK DEMO FLOW

  1. Landing Page (/) - Show hardware, stats, architecture
  2. Command Center (/command) - Live heatmap, Comms tab for AEGIS chat
  3. Fleet Tab - Click "View Live Feed" to show SENTINEL modal
  4. Overview Tab - Show Simulation Control and risk predictions
  5. Back to Landing - Hardware section for closing

🎤 KEY TALKING POINTS

  • "Predictive, not reactive" - Say this multiple times
  • "100% local" - Emphasize data sovereignty
  • "30+ stressor signals" - Shows depth of model
  • "Sub-10ms inference" - Technical credibility
  • "DGX Spark + Nemotron" - NVIDIA integration points
  • "Running live, right now" - Not vaporware

Prepared for SparkHacks 2026