Technical complexity
Equipment, fuels, and capacity require controlled knowledge and clear boundaries.
RAG sales agent · Conversational experience and integrations
Conversation is only the interface. Laura turns intent into commercial action.
We built a RAG agent for Sigmart Group's boiler and steam division that retrieves technical knowledge, recognizes context, records opportunities, and gives sales a clear next action.
Sigmart Group · Industrial equipment and servicesWeb agent, technical knowledge, CRM, and sales activation
01Typing in real time
24/7inquiry capture and guidance
1context across conversation and CRM
7steps in the operating loop
Livehandoff to the sales team
* Capabilities reflect the delivered scope. Commercial impact should be measured with post-launch operating and conversion data.
The opportunity
An industrial inquiry rarely fits inside a form.
A conversation may begin with capacity, fuel, efficiency, or an urgent failure. For sales to act, it needs the product, company, project size, history, and urgency—not only a name and email.
Every lead could lose momentum before reaching a specialist:
- Technical inquiries without immediate guidance
- Contact details without project context
- History separated from the conversation
- Follow-up dependent on manual capture
The challenge
Connecting natural language to real commercial work.
The solution had to converse with rigor, avoid fabricated answers, and operate across existing systems without giving the model direct access to data or sensitive actions.
Fragmented context
The website conversation and commercial history lived in different places.
Incomplete handoff
An alert without product, urgency, and summary forces sales to reconstruct the need.
The system
An agent between the customer, knowledge, and sales.
Laura listens first, retrieves evidence when needed, completes only missing information, and runs authorized tools to leave the opportunity ready for human follow-up.
Listen
Interprets the need expressed in natural language.
Investigate
Retrieves technical knowledge and history when appropriate.
Qualify
Structures product, industry, size, contact, and urgency.
Activate
Updates the CRM and alerts the team with an actionable brief.
Not a chatbot
Answering is a capability. Deciding and acting is the agent's job.
A conventional chatbot ends after delivering text. Laura maintains state, selects tools, and connects the conversation to an operating consequence—always within defined permissions and rules.
Traditional chatbotLaura · AI agent
Follows a script or FAQReasons about intent, context, and missing data
Starts every chat from scratchCan retrieve the customer and previous interactions
Delivers isolated answersDecides when to retrieve knowledge or use a tool
Captures data without a destinationCreates or updates customers and opportunities in the CRM
Ends when the window closesActivates the next step with the sales team
Retrieval-Augmented Generation
Laura is a RAG agent: it retrieves evidence before generating an answer.
For technical inquiries, Laura does not rely only on what the model learned during training. It first searches a controlled knowledge base, adds the relevant passages to context, and then answers or decides to escalate to a specialist.
01Customer question
02Knowledge retrieval
03Relevant technical context
04Grounded answer or escalation
Less improvisation
Reduces the risk of fabricated answers about equipment, fuels, and capacity.
Updatable knowledge
The company can improve sources without retraining the model from scratch.
Better qualification
The conversation combines customer needs with technical context useful to sales.
Perceive
Understands intent, facts, and urgency signals.
Decide
Chooses whether to ask, retrieve, or execute.
Act
Records, notifies, and confirms the next step.
Laura does not invent prices or replace the specialist's technical judgment.
Capabilities
A conversational experience connected to operations.
The interface is simple for visitors; underneath, the agent coordinates knowledge, memory, CRM, and internal communication.
Technical guidance
Retrieves equipment, capacity, fuel, and service information before answering.
Useful memory
Maintains the session and retrieves commercial context after the user identifies themselves.
Progressive qualification
Requests only missing information, one question at a time.
Connected CRM
Creates or updates customers and records each request as an independent opportunity.
Sales-ready handoff
Sends Google Chat a brief with contact, product, size, urgency, and summary.
Local expectations
Communicates the next step based on business hours in El Salvador.
Operating change
From contact details without context to a sales-ready opportunity.
Passive captureWith Laura
The same form for every needA conversation adapted to intent
Contact details without technical contextStructured product, industry, size, and urgency
History far from the conversationCommercial context retrieved when appropriate
Manual recording and notificationCRM and alert activated from the same flow
Visitors do not know what comes nextA follow-up expectation is confirmed
Commercial continuity
always_on.salesFewer opportunities fall between a conversation and follow-up.
Laura reduces the risk of losing leads to incomplete information, delay, or missing records. Every qualified conversation can become a traceable CRM object and reach the channel where sales already works.
01Conversation
02Qualified need
03Customer + opportunity
04Sales alert
05Human follow-up
capture outside business hours
The inquiry is guided and recorded for the next shift.
context ready to prioritize
Sales receives product, company, size, urgency, and summary.
every request retains its source
One customer can generate separate opportunities without losing history.
measurement.readyKPIs prepared for the next measurement stage
- Conversation → lead
- Time to first human contact
- Record completeness
- Conversion to quote
* Expected operating benefits based on implemented capabilities; these are not audited conversion results.
Architecture
Reasoning separated from system access.
Claude interprets and selects tools. Node.js and Express validate actions; Supabase/PostgreSQL stores knowledge, customers, and opportunities; Google Chat activates the team, and Railway runs the service.
ClaudeNode.jsExpressSupabasePostgreSQLFull-text searchGoogle ChatRailway
Agency with boundaries
- Explicit tools with input schemas
- The model has no direct database or webhook access
- Each customer can only access their own context, never another user's private data
- Controlled knowledge before technical answers
- Rules for pricing, out-of-scope topics, and unknown information
Orbi Labs
Does your website receive conversations that never become follow-up?
We design agents that understand the business, work with your systems, and leave your team a concrete action.
