-
-
Thumbnail
-
-
-
-
Inspiration
Selling used items on campus or community markets can be very draining. From people demanding a lower price to constantly arguing for a few dollars to suddenly ignoring you while still in the middle of a conversation, selling used items can feel like a part-time job.
I wanted to create an easy, automated way for me (and all other students who want to earn some extra money by selling used items) to do so without being constantly on their phones. With Campus Agent, students can use an automated AI "employee" to handle the negotiation strictly within their limits as they can focus on their other priorities and close the sale themselves.
What it does
Campus Agent: Autonomous AI Bargaining Engine is an end-to-end negotiation and instant settlement platform:
- Intelligent Bargaining: Buyers chat with an AI negotiation agent configured with private seller constraints (target price, strict reserve floor, and custom seller persona traits).
- Firm Reserve Defense: The agent dynamically resists lowball offers by articulating product value, yet smoothly compromises when realistic counter-offers are made.
- Instant Deal Detection & Settlement: The moment an agreement is reached, the system generates an integrated PayPal Sandbox checkout link for instant escrow payment.
- Automated Workflow Orchestration: Finalized transactions trigger Zapier webhooks to immediately record the deal, notify external endpoints, and streamline order fulfillment.
How we built it
Built as a solo developer sprint from architecture to deployment:
- Core Reasoning Engine: Powered by Google Gemini through the Gemini API, leveraging targeted system instructions and dynamic prompt formatting to calculate compromise curves.
- Frontend & State Management: Built entirely in Python using Streamlit, managing dynamic conversational memory and reactive UI cards.
- Payment & Automation Rails: Integrated PayPal REST APIs for sandbox checkout alongside Zapier Webhooks for post-deal dispatch.
- Engineering Discipline: Managed iteration, feature rollbacks, and component reliability via strict Git version control to keep dependencies clean.
Challenges we ran into
- Prompt Rigidity vs. Negotiation Flexibility: LLMs naturally lean toward being overly polite or giving in to persistent buyers. Engineering a prompt contract where Gemini firmly protected a private reserve price—without sounding robotic or stubborn—required extensive prompt tuning and edge-case testing.
- Deterministic State Transition in a Non-Deterministic Chat: Ensuring the model reliably emitted structured signals when a deal was legally "closed" (to avoid false-positive payment triggers) required strict output schema alignment between Gemini and the Streamlit state handler.
- Rapid Full-Stack Scoping: Balancing UI styling, real-time API latency, and external webhooks within a tight hackathon timeline as a solo builder required ruthlessly cutting non-essential frontend refactors and focusing strictly on the critical path.
Accomplishments that we're proud of
True End-to-End Autonomy: Created a functional system that takes a text input from the buyer through to determining a price, automatically creating a PayPal payment and using Zapier to complete the task – entirely without human intervention
High Negotiation Fidelity: Correctly negotiated the price down from the listed price and successfully held a determined $85.00 minimum when prompted by a buyer
Zero-Dependency Architectural Stability: Maintained a light, stable and reproducible codebase without relying on additional UI weight
What we learned
- Real-world AI agents are only as good as their guardrails: explicit fallback instructions and mathematical boundaries are critical when money is on the line.
- Webhook-driven micro-automations (like Zapier) unlock enterprise-level post-transaction pipelines in minutes when integrated cleanly with modern LLMs.
What's next for Campus Agent: Autonomous AI Bargaining Engine
- Multi-Item Inventory Support: Allowing campus sellers to batch-upload items and manage a unified fleet of negotiating agents.
- On-Chain & Multi-Rail Payments: Expanding beyond PayPal to support Venmo, Cash App, and stablecoin micro-settlements.
- Dynamic Market Price Intelligence: Enabling agents to pull real-time historical pricing from eBay and local listings to auto-suggest optimal reserve floors for students.
