Coco

Inspiration

Small business owners deal with customer inquiries every day, but a lot of their time gets lost in repetitive operational work: recording leads, understanding which customers are serious, remembering follow-ups, and figuring out what needs attention first.

I wanted to build an AI assistant that goes beyond simply answering questions. Instead of making the owner manage every step manually, Coco turns natural-language customer inquiries into structured business actions. That idea became Coco — an AI operations agent for small businesses.

What it does

Coco helps a business owner manage customer inquiries and turn them into organized actions. A business owner can simply tell Coco something like:

"Rahul wants catering for 200 people tomorrow and is ready to book." Coco can understand the request, create the lead, analyze its importance, determine its priority, recommend the next action, manage follow-ups, and surface leads that need attention.

Coco currently supports:

• Creating customer leads from natural language
• Retrieving existing leads
• Updating lead status
• Analyzing lead intent and priority
• Identifying hot, warm, and cold leads
• Recommending next actions
• Managing follow-up information
• Identifying what needs the business owner's attention The goal is to reduce repetitive operational work while keeping the business owner in control of important decisions.

How we built it

Coco was built in Python using Amazon Bedrock, Amazon Nova 2 Lite, and Strands Agents. The agent is connected to business tools that allow it to interact with a SQLite database instead of only generating text. The main architecture is: User → Amazon Nova 2 Lite → Strands Agent → Tools → SQLite → Business ActionWe created tools for operations such as creating, retrieving, and updating leads, analyzing lead priority, managing follow-ups, and identifying important actions.

This tool-based architecture allows Coco to reason about a request and then perform an actual operation on business data.

Challenges we ran into

One of the biggest challenges was working with real AWS infrastructure rather than only prototyping with an AI API.

The Claude models we initially planned to use were not available for our AWS account, so we adapted the project to use Amazon Nova 2 Lite. We also encountered an Amazon Bedrock throughput limitation where the model required an inference profile instead of direct on-demand invocation. We had to configure the correct inference profile and resolve AWS authentication issues during development. Another challenge was designing the agent so that the AI could safely interact with structured business data through tools instead of simply responding with text.

Accomplishments that we're proud of

We are proud that Coco became a working AI agent rather than just a chatbot prototype. The agent can take an unstructured customer request and turn it into a real business workflow: storing the lead, analyzing its priority, updating its status, and recommending what should happen next. We are especially proud of the lead intelligence workflow, because it allows the business owner to quickly understand which opportunities deserve immediate attention. We also successfully integrated Amazon Bedrock and Nova 2 Lite with Strands Agents and connected the agent to a persistent SQLite database.

What we learned

We learned that building an effective AI agent is not just about choosing a powerful language model. The model becomes much more useful when it has clearly defined tools, structured data, and rules for taking actions. We also learned a lot about AWS Bedrock, inference profiles, authentication, agent tool calling, SQLite, and designing workflows around real business operations. Most importantly, we learned to think about AI as an operator that can take action, rather than just an assistant that produces text.

What's next for Coco

The next step is to make Coco increasingly autonomous. We want Coco to monitor routine business operations in the background, identify important changes, prepare follow-ups, and surface only the situations that genuinely require the owner's attention. Over time, Coco could expand beyond lead management into areas such as appointments, customer communication, payments, inventory, and daily business planning.

The long-term vision is simple: Coco handles the repetitive operational work. The business owner focuses on the decisions that actually matter.

Built With

• amazon-bedrock
• amazon-nova-2-lite
• amazon-web-services
• python
• sqlite
• strands-agents