ProjectNudge - Fast, Fluid, & Functional: Guided Build Night w/ LiquidMetal AI
AI Tinkerers - Seattle
Hackathon Showcase

ProjectNudge

Team consisting of MSR Senior Researcher (Stanford PhD; Open RAN/Azure), Groundlight Applied Scientist (VLMs), Juniper SDE‑II (C/C++ routing), and Identity Digital PM (AI product).

4 members

Project Description: AI Smart-Bucket Collections System (ASBCS)

The problem solved is the high cost, inefficiency, and poor customer experience associated with traditional, human-run collections agencies. Insurance companies lose revenue and goodwill due to reactive, one-size-fits-all debt recovery processes.

Nudge is an autonomous, AI-driven debt collection and triage system. It uses the Gemini API as a sophisticated decision-making engine to replace human collections agents, providing a strategic, channel-specific action plan for every past-due account.

Core Features:

Smart-Bucket Triage: An LLM analyzes account data (Amount, DPD, Name, Phone, Email) and, crucially, the conversation history (status). It dynamically triages accounts, prioritizing based on value/urgency and contextualizing based on past success/failure.

Structured Action Planning: The LLM outputs a single, parseable JSON action plan that dictates the next_action (EMAIL, TEXT, CALL, MANUAL_REVIEW), suggested_time, and a precise strategy.

Multi-Channel Automation: Integrates with Twilio to execute the determined action instantly, communicating via the customer’s preferred or last-successful channel.

Real-Time Context: Conversation status is stored and updated after every automated action, ensuring the LLM’s next decision is always based on the latest customer interaction (e.g., respecting a payment promise).

Judging Criteria

Creativity: We use the LLM not for conversational chat, but as a complex, dynamic Triage Engine guided by a comprehensive system prompt. This shifts the AI from a simple chatbot to a proactive, rules-based strategist, drastically improving operational predictability and reliability.

Reliability: Reliability is ensured via the structured JSON output format dictated in the system prompt. This prevents LLM “hallucinations” from breaking the downstream automation pipeline (Twilio integration). All state is managed via Firestore, providing fault tolerance and real-time synchronization.

Real-World Impact: ASBCS offers insurance companies a high-impact, immediate solution for reducing collections agency fees (typically 15-30% of recovered debt) while improving customer retention through empathetic, non-confrontational, and context-aware communications.

Technology Stack

LLM/Decision Engine: OpenAI API

Communication: Twilio (SMS and Call actioning)

Data/Storage: Google Firestore (Real-time account status and history)

Frontend/UI: HTML, Tailwind CSS, JavaScript (for the operations dashboard demo)

Anthropic LiquidMetal AI OpenAI API Twilio (Multi-channel communication/actioning)