Support Assistance

Support Assistant: an AI that becomes your business’s support team. It understands your brand, enforces policies, verifies requests, and scales with you - instantly, without hiring.

Comment

Inspiration

It started with a simple observation at a local bakery. As businesses grow, their customer messages explode and very quickly it becomes chaos. We saw small business owners overwhelmed, relying on personal WhatsApp numbers, manual replies, and sticky notes. They were losing time and missing customers, not because they didn’t care, but because they lacked the infrastructure.

We realized there was a massive gap. You either have "no system" or you buy expensive, complex enterprise software like Salesforce. We wanted to build something in the middle. A tool that gives a solo founder the support infrastructure of a Fortune 500 company, instantly.

What it does

Support Assistance is an intelligent customer service platform powered by Gemini 3. It isn't just a chatbot; it is a policy-enforcing agent.

Here is the flow:

  • Instant Onboarding: A business owner inputs their brand name, uploads their policy documents, and defines their tone and feel of the user interface.

  • The Generation: In seconds, the system spins up a unique, dedicated support instance with a shareable link and QR code.

  • The Customer Experience: The AI combines deep contextual reasoning with multimodal image analysis. It evaluates each request against business policies while examining uploaded photos for authenticity, enabling automated, fraud-aware decisions powered by the full capabilities of Gemini 3.

  • The Admin Command Center: We built a secure Admin Portal where owners can log in (via username/password) to oversee the operation. They can view full chat logs, track active sessions, and intervene if necessary.

How we built it

We architected Support Assistance as a scalable platform rather than a simple wrapper.

Backend: We used FastAPI to handle the heavy lifting, managing a SQL database that stores business profiles, chat logs, and authentication data.

The Brain: The core logic is powered by Google Gemini 3. We utilize its massive context window to hold dynamic business policies in memory during conversations.

Frontend: The user interface (both the customer chat and the Admin Portal) is designed to be minimalist and responsive, ensuring it looks professional on both mobile and desktop.

Dynamic Prompting: We engineered a "meta-prompting" system. When a business onboards, we don't just save their text; we compile it into a structured System Instruction set that dictates how Gemini behaves for that specific tenant.

Gemini 3 Integration (Key Features)

We leveraged Gemini 3 not just for text generation, but for its reasoning and instruction-following capabilities.

  • Policy Reasoning: Standard LLMs answer questions. Gemini 3 adjudicates. If a customer asks for a refund on a "final sale" item, Gemini analyzes the stored policy context and politely refuses, explaining why based on the specific business rules.

  • Multimodal Forgery Detection: We utilize Gemini’s advanced vision capabilities to verify claims. If a user uploads a photo of a "damaged product," Gemini analyzes the image metadata and visual artifacts to determine if the image is genuine or if it has been edited/manipulated, flagging potential fraud before a human even sees it.

  • Tone Adaptation: We use Gemini's fine-grained control to swap personas instantly, from a professional law firm assistant to a casual surf shop buddy, based on the onboarding data.

Challenges we ran into

The hardest part was the "Hallucination vs. Policy" balance. Early versions of the AI were too nice; they would grant refunds that violated company policy just to be helpful! We had to spend significant time prompt-engineering constraints to ensure the AI understood that following the rules was more important than pleasing the user. Additionally, building a secure authentication flow for the Admin Portal while keeping the customer side frictionless required careful architectural planning.

Accomplishments that we're proud of

  • The Support: We successfully built a system where the AI acts as a knowledgeable representative, not a generic bot.

  • The Admin Dashboard: Creating a secure, real-time view for business owners to track their AI's performance feels like a game-changer for trust.

  • Zero-Code Deployment: We are proud that a non-technical baker or mechanic can set this up in under 3 minutes and have a working support link.

What we learned

We learned that AI is most powerful when it’s grounded in strict constraints. Language capability alone isn’t enough for business; systems need structure, clarity, and boundaries. We also learned that for small businesses, the UI/UX is just as important as the model. If the Admin Portal isn't intuitive, they won't trust the AI.

What's next for Support Assistance

We plan to turn this into a full CRM.

  • Payment Integration: Allowing the AI to process payments or issue refunds directly.

  • Voice Support: Using Gemini's capabilities to allow voice support.

  • Plugin System: One-click integration with Gmail, Discord and WhatsApp Business API.

Built With

Share this project:

Updates