Retrieval Augmented Generation

Retrieval augmented generation (RAG) is a popular way to ground prompt requests to large language models (LLMs). Grounding adds domain-specific knowledge or customer information to the prompt, giving the LLM context to respond more accurately to a question or task.

 To break it down, RAG:

• Retrieves relevant information from a knowledge store containing structured and unstructured content.
• Augments the prompt by combining this information with the original prompt.
• With the augmented prompt, the LLM generates a response.

Many LLMs are trained generally across the Internet on static and publicly available content. RAG adds domain-specific information to help LLMs give you better responses to your prompts. With RAG, you can extract valuable information from all sorts of content, such as service replies, cases, knowledge articles, conversation transcripts, RFP (request for proposal) responses, emails, meeting notes, frequently asked questions (FAQs), and more.

AI grounding

Grounding refers to the process of connecting an AI model’s output, particularly from generative AI models like Large Language Models (LLMs), to verifiable and real-world sources of information. AI models are born generalists. To get smarter responses and the ability to complete complex tasks, you need to provide your AI model with the right information. Grounding gives ‌LLM domain-specific knowledge and customer information. Without grounding, AI is like someone who has memorized a dictionary but never experienced the real world. With grounding, AI can:

• Recognize objects in photos
• Understand human instructions better
• Interact in real-life situations (like a robot)

Example- Without Grounding, AI guesses “dog” as an animal, just based on text, but with grounding, AI sees images of dogs and understands what they look like.

 

Why is AI grounding critical?

• Reducing Hallucinations - Hallucinations occur when an AI model generates the output that sound correct but is actually false or not based on real data. For example, you ask, who is the president of Mars? And if AI replies like “John is the current president of Mars” this is called Hallucinations as it made a fake outcome. AI grounding reduce “Hallucinations”.
• Enhancing Accuracy and Relevance: Grounded AI models can provide more accurate, relevant and contextually appropriate response.
• Providing a auditability and trust: Grounding allows for greater transparency and auditability. When an AI response is grounded, it can often provide links or references to the sources it used, enabling users to verify the information and building trust in the AI's outputs.
• Customization and Specialization: For businesses and specific applications, grounding is essential for tailoring AI responses to unique datasets. For example, a customer support chatbot can be grounded in a company's knowledge base to provide accurate product details or troubleshooting steps specific to that company.

How is AI Grounding Achieved (Common Techniques)?

• Retrieval-Augmented Generation (RAG): This is a popular technique where the AI model retrieves relevant information from a curated database or external knowledge base before generating a response. This retrieved information then "grounds" the model's output.
• Connecting to Search Engines: Grounding can involve giving AI models access to real-time information from the web through search APIs (like Google Search).
• Integrating with Proprietary Databases: For enterprise applications, AI models can be grounded by connecting them to internal company data, documents, or knowledge graphs.
• Sensorimotor Grounding (Embodied AI): In robotics and embodied AI, grounding involves connecting language understanding with real-world sensory inputs (e.g., cameras, microphones) and physical actions. This helps robots understand and interact with their environment.
• In-context learning: Feeding relevant data directly into the prompt can also serve as a form of grounding, guiding the LLM's response based on the provided context.

Health Cloud Data Model

• Healthcare systems, such as EHRs, lab systems, and billing systems, are often built by different vendors and require a common way to communicate. HL7 provides that standard format and protocol.  
• Salesforce Health Cloud offers a comprehensive set of objects and fields that support various healthcare business processes. This model is built to represent patient, caregiver, clinical data, care plans, and provider relationships in a way that aligns with real-world healthcare workflow and standards like HL7 and FHIR.

Throughout a patient’s journey, various processes and events take place—such as scheduling appointments, receiving lab results, using medical devices during treatment, getting prescriptions, and processing insurance claims and so on. To support these processes, Salesforce provides different kind of data models that are linked together to reflect the complete healthcare experience such as:

Data Model	Description
Household Data Model	The household data model maps the relationships between patients or members and the people and organizations who participate in their care.
Clinical Data Model	The clinical data model aligns with FHIR and supports many of the HL7 message types. This data model captures information from an Electronic Health Record (EHR) system via integration middleware. It displays information commonly associated with EHRs such as patient or member encounters, chronic conditions, allergies, medications, and immunizations.
Health Insurance and Claims Data Model	The Health Insurance and Claims data model provides objects for managing the ways healthcare is paid for. These objects represent information about members and their employers, insurance coverage, and dependents. This data model includes objects for claims submitted to a member’s health insurance company by an entity or person who provides a service, medication, or device to the member.
Care Program Data Model	A Care Program is like a blueprint for treating patients who share a common condition or need such as diabetes, heath disease, post-surgery recovery or behavioral health. It includes Goals & care plans, Task and interventions, assessment, care team coordination, and patient enrollment. The Care Program data model helps you capture those activities so you can use Health Cloud to plan and track them better.
Integrated Care Management Data Model	Provides objects to store clinical data related to a patient or member’s care plans. This data model is USCDI and FHIR R4-aligned, which helps with your system’s interoperability.
Social Determinants of Health Data Model	Social Determinants of Health are non-medical factors that affect a person's overall health, like (Housing, Food security Transportation, Employment, Education Social support. These are critical to understanding and improving a patient’s health, especially for care teams managing chronic conditions or vulnerable populations. Salesforce Health Cloud helps care teams track non-medical factors like housing, food, and transportation that affect a patient’s health. It uses assessments to identify risks, creates care plans and referrals to connect patients with community services, and helps improve outcomes by addressing the whole person — not just the medical side.
Provider Data Model	The Provider data model represents healthcare providers, including individual practitioners and facilities, the network they participate in, and their specialties and credentials.
Intelligent Appointment Management Data Model	Provides objects to help you work with multiple scheduling source systems operating on different electronic health record platforms.
Utilization Management	Utilization Management (UM) is the process of reviewing and approving healthcare services to make sure they are medically necessary, appropriate, and covered by insurance before treatment is given. It includes objects like Care Request, CareRequestItem, CareRequestReviewer, etc.
Benefits Verification	Supports providers, payers, and life sciences organizations in determining patients’ insurance coverage and the benefits they are entitled to, especially for specific procedures, visits or prescriptions. It includes objects like Member (Person account), Payer (Person Account), Case (BV), Member Plan, Coverage Benefit etc.

Healthcare Industry Sectors

All five healthcare industry sectors contribute to better patient. 

• Providers
• Payers
• Pharmaceuticals
• MedTech
• Public Sector health

Who Are the Providers?

• Providers include the people who take care of patients and the facilities where patients are cared for. People providers are who you think they are—doctors, nurses, physical therapists, and other clinical staff. These professionals provide preventive, curative, or rehabilitative healthcare services.
• Provider facilities are things like hospitals, clinics, and skilled nursing facilities.

Who Are the Payers?

In healthcare, a payer is an entity responsible for covering the cost of medical services provided to patients by healthcare professionals (known as providers). They are the ones who pay bills for healthcare. Payers call their customers “members”. Payers do the following tasks.

• Offer health plans/coverage
• Collect premiums – They receive money from individuals, employers, or government sources to fund these plans.
• Process claims- When a patient receives care, the provider sends a claim to the payer, and then reviews and processes.
•  Engage members in wellness and care management programs.

Type of Payer-

• Major National Plans
• Regional and Local Plans
• Provider-Led Plans
• Government Plans

What are government payers?

Government/Public Payers- These are the programs funded by federal or state government to provide healthcare coverage to specific populations:

• Medicare - Primarily for individuals 65 and older and certain younger people with disability or end-stage renal disease
• Medicaid – Provides health coverage to low-income individuals and families and those with certain disabilities. It's jointly funded by the federal and state.
• CHIP (Children’s Health Insurance Program) – Provides low-cost health coverage to children in families who earn too much to qualify for Medicaid but cannot afford private insurance.
• TRICARE- Primarily for active-duty and retired uniformed officers and their families.
• Veterans Health Administration (VA)- Provides coverage to eligible veterans.

Who are the Pharmaceuticals?

Pharmaceutical manufacturers belong to a sector of the healthcare industry known as life sciences. In general, life sciences focus on research and manufacturing of drugs and devices used in healthcare. Life sciences include biotechnology experts, medical device developers, and pharmaceutical manufacturers.

A drug, also called a medication, goes through a long lifecycle before it gets on the market. Here are the stages it travels

1. Research and Development (R&D): Research is the scientific investigative undertaking that results in the discovery of a new drug. Development is when these findings are used to produce the drug, which is then tested in clinical trials.
2. Clinical trials: Carefully conducted clinical trials are performed on volunteers to determine if a treatment works, is safe, and is more effective than an existing treatment.
3. Regulatory review: In the US, findings from clinical trials are submitted for review by the Food and Drug Administration (FDA) to approve or deny the drug for sale.
4. Commercial marketing and sales: A new drug launches on the market with marketing campaigns targeted to physicians and patients, while pharma sales reps educate physicians on the new drug and how to prescribe it.

Pharmaceutical Products- If we were to list all the drugs on the market today, we’d need a while, because there are thousands! Instead, let’s talk about the different drug classifications. 

• Generic Vs Brand Name
• Brand Name Drugs -These are original drugs developed, tested, and marketed by pharmaceutical companies under a specific trade name. Example: Tylenol is a Brand name that has the Active Ingredient (Generic Name): Acetaminophen. Protected by patents for several years, meaning no other company can sell the same drug under a different name.
• Generic Drugs-These are copies of brand name drugs with the same active ingredients, strength, dosage form, and route of administration. Approval: Approved by regulatory bodies (like the FDA) only after the patent of the brand drug expires.
• Small Molecule Vs Biologic
• Small-molecule drugs are those manufactured in a lab by mixing different chemicals until the right compound is formed. Small-molecule drugs treat common conditions like infections, high blood pressure, diabetes, and high cholesterol. 
• Biologics, on the other hand, are much more complex. They’re discovered and manufactured in living organisms, like plant and animal cells. They’re used for conditions like cancer, multiple sclerosis (MS), and hepatitis C. There’s more to know, but that’s it for now.

Pharmaceutical Customers-There are four main customers in the pharma sector.

• Physicians prescribe the drugs to patients.
• Payers pay for most of the prescription cost—the patient pays the other portion in the form of a co-pay.
• Patients take the drugs or receive the treatments.
• Distributors store and distribute the drugs.

Who is the MedTech?

Medical devices and technology (MedTech) fits into the life sciences sector of the healthcare industry. Life sciences focus on the research and manufacturing of drugs and devices used in healthcare.

MedTech produces the familiar and the futuristic. MedTech companies include those that manufacture devices, like bandages or robotics, and those that provide services, like software to translate x-rays into images.

MedTech Customers

Many of the healthcare sectors serve the same customers, so you may be familiar with this list. Medtech customers include: 

• Payers
• Providers
• Patients
• Distributors

Who is Public Sector Health?

The public sector health system refers to healthcare services provided, funded, or regulated by the government. It aims to ensure affordable, accessible, and equitable healthcare for all citizens, especially the poor and vulnerable.

 

AgentForce

What is AgentForce?

Salesforce Agentforce is an AI-powered platform designed to create and deploy digital agents that can automate tasks, improve productivity, and enhance customer service. 

Building Blocks of Agents-

Agents-

• These are the core AI assistants that perform tasks and interact with users.
• They are more autonomous than other conversational AI solutions, able to identify opportunities for action and initiate tasks within defined parameters
• An agent can perform business tasks on behalf of the users.

Topics –

• Topics define the scope of an agent's capabilities and categorize the tasks it can handle.
• They guide the agent in understanding user requests and applying the appropriate actions and instructions. For example, a topic could be "Account Management" or "Order Status Lookup".

Instructions

• Instructions provide guidance to the agent on how to execute actions and interact with users.
• They define the agent's decision-making process and ensure consistent and appropriate responses.

Actions

• Actions are the specific tasks an agent can perform, such as looking up order status, updating account information, or generating a return label.
• These can be standard Salesforce actions or custom actions built for specific needs.

Reasoning Engine

• The reasoning engine orchestrates the interaction between topics, actions, and the Large Language Model (LLM)
• It determines which topics and actions are relevant to a user's request and guides the agent's actions accordingly
• Agentforce is using Atlas Reasoning Engine

Large Language Model (LLM)

• The LLM enables the agent to understand natural language, communicate effectively with users, and generate relevant responses.

Security

• Einstein Trust Layer

What are the Five Attributes of an Agent?

• Role
• It's similar to the roles that you've already defined today across your Customer 360. 
• For example, if you're using Sales Cloud, you've defined roles for your SDRs and your account executives - their metrics and dashboards.

• Data
• All of this data is already in the Salesforce Platform.
• Structured data like your custom fields as well as unstructured data like Slack conversations and knowledge articles. 
• With our new Data Cloud Zero Copy Partner network, you can also pull in all of your external data lakes and warehouses to be activated by Agentforce. 

• Actions
• When you think about what workflows you want your Agent to be able to perform and automate.
• These are the actions you've already built - the business processes into the Salesforce platform - every flow, every line of APEX code, every API integration, it’s all available for you to point and click and empower your Agentforce.

• Channels
• Thanks to the Salesforce platform, Agentforce has access to every digital channel natively within Salesforce whether that's Slack, Whatsapp, or your company website. 

• Guardrails and Trust 
• Provide an Agent with guardrails around what they can and cannot do. 

 What is Einstein Trust Layer?

AI agents are integrated with the Einstein Trust Layer, which is a secure AI architecture natively built into Salesforce. Designed for enterprise security standards, the Trust Layer lets you benefit from generative AI without compromising your customer data. It also lets you use trusted data to improve generative AI responses.

• Data grounding: The Trust Layer ensures that generative prompts are grounded and enriched in trusted company data.
• Zero-data retention: Your data is never retained by a third-party LLM provider.
• Toxicity detection: Potentially harmful LLM responses are detected and flagged.
• AI monitoring: AI interactions are captured in event logs, giving you visibility into the results of each user interaction.

 What are AI Guardrails?

AI guardrails are mechanisms that ensure your AI project operates legally and ethically. You need guardrails to prevent AI from causing harm with biased decisions, toxic language, and exposed data. Guardrails are necessary to protect your project from technical attacks.
There are three types of AI guardrails—security guardrails, technical guardrails, and ethical guardrails.

• Security Guardrails-These guardrails ensure that the project complies with laws and regulations, and that private data and human rights are protected. Common tools here include secure data retrieval, data masking, and zero-data retention. Secure data retrieval means that your project only accesses data that the executing user is authorized to access. For example, if a person with no access to financial records triggers a response from the AI model, the model shouldn’t retrieve data related to financial records. Data masking means replacing sensitive data with placeholder data before exposing it to external models. This ensures that sensitive data isn’t at risk of being leaked. Model providers enforce zero-data retention policies, which means that data isn’t stored beyond the immediate needs of the task. So, after a response is generated, the data used is no longer available.
• Technical Guardrails-These guardrails protect the project from technical attacks by hackers, such as prompt injection and jailbreaking, or other methods to force the model to expose sensitive information. Cyberattacks can cause your project to generate untrue or harmful responses.
•  Ethical Guardrails-These guardrails keep your project aligned with human values. This includes screening for toxicity and bias. Toxicity is when an AI model generates hateful, abusive, and profane (HAP) or obscene content. Bias is when AI reflects harmful stereotypes, such as racial or gender stereotypes. As you can imagine, that’s a disaster! Since AI learns its responses, toxicity and bias could be a sign that your data is introducing unwanted language and ideas to your model. Toxicity detection identifies responses that might have toxic language, so you can review it manually and make adjustments to reduce toxicity.

What is Agentforce Data Library?

Agentforce Data Library is a Salesforce feature designed to enhance the accuracy and relevance of AI agents' responses by providing them with access to an organization's trusted data. It uses "grounding" to index various data sources, allowing AI agents to base their answers on accurate and specific information.
Here's a breakdown of what it is and how it works:

What it is:

A centralized repository for structured and unstructured data, including

•  Knowledge articles
• File uploads (e.g., PDFs, text, HTML)
• Web sources
• Salesforce fields and records

How it works:

·       Data Ingestion: You upload or connect your data sources to the Data Library.

•  Indexing: The Data Library indexes this content, making it searchable for AI agents. This process also automates several configuration steps in Salesforce Data Cloud and Prompt Builder, such as creating data streams, mapping data objects, and setting up search indexes and retrievers.
• Grounding: When an AI agent receives a query, it uses the indexed data to "ground" its response. This means it retrieves the most relevant information from your specific data library to ensure its answers are accurate, contextual, and aligned with your organization's policies and information.
• Generative AI Integration: The retrieved information is then combined with Generative AI to formulate a clear, helpful, and natural-language response.

Agentforce for Sales

• Agentforce SRD
• In sales, an SDR (Sales Development Representative) is an entry-level role focused on generating and qualifying leads for a sales team. They are the first point of contact for potential customers, engaging with them through various channels like cold calling, emailing, and social media to identify if they are a good fit for the company's products or services. SDRs do not close deals; instead, they pass qualified leads to sales representatives
• Agentforce SDR, also known as AI SDR agents, helps your team nurture leads at scale 24/7. SDR agents are proactive, autonomous agents that send personalized outreach, answer your customers’ questions accurately, and even book meetings on your behalf. This is transformative for sales teams and allows them to focus on building customer relationships and pipeline growth. Here, you learn more about these agents.
• Agentforce Sales Coach
• Sales teams continually face new challenges. A sales rep leaves, and the team scrambles to cover until a new hire is up to speed. An inexperienced sales rep needs support for a negotiation with a new customer they’re struggling with. Managers just don’t have the time to review every deal and give every seller the 1:1 coaching they need.Good news: Salesforce now offers a solution that helps managers and sales reps alike. Sales coaching in Agentforce is like having a personal trainer in Salesforce. 
• Agentforce Sales Coach provides deal-specific, personalized feedback to sellers on their messaging and customer communication. It provides
• Autonomous, on-demand coaching: Reps can receive guidance anytime, without needing a live coach
• Pitch practice before meetings: Sellers can rehearse their pitch and get tailored, actionable feedback based on the deal and its current stage
• AI-driven role-play scenarios: Reps can practice tough conversations with a simulated “customer” or “buyer.
• Personalized feedback on role-plays: Reps can improve their approach using insights specific to their deal.

 

Prompt Builder

 What is Prompt Builder?

Before we understand about Prompt Builder. It would be good to know what LLM is and how can we utilize in our business.

A Large Language Model (LLM) is an advanced AI system trained on huge amounts of text to understand and generate human-like language. It can answer questions, write content, translate, summarize, and more—just like ChatGPT. LLM can make a big impact on business. For example, they can help you to generate the personalized email to customers or provide case summary in a glance and many more. 

However, LLMs cannot generate great content themselves. They need some guidance or instruction based on which LLMs can create the output. These instructions are called a Prompt. In short, A Prompt is a specific instruction or question given to an LLMs to generate its output. These reusable, predefined instructions (“Prompt”) are collectively referred to as the Prompt Template. The Prompt Template has merge fields from salesforce record.

A Prompt Builder is a tool that helps users create, manage and deploy prompts for LLs.

Key Features of Prompt Builder-

1. Drag and drop interface – Build prompts without writing code using a simple interface.
2. Template Creation- Prompt Builder allows the user to create a reusable prompt template.
3. Integration with business data- A significant advantage of Salesforce's Prompt Builder is its ability to integrate directly with existing business data (e.g., customer information, product details). This "grounds" the AI's responses, making them highly relevant and personalized rather than generic.
4. Security and trust- When you generate a response from a prompt built in Prompt Builder, the prompt is sent to the Einstein Trust Layer, which masks any sensitive data before sending it to the LLM. When prompts are sent to external models through the shared trust boundary, your data is encrypted to ensure its security in transit. Additionally, any sensitive information within the prompts is masked. Finally, based on the “zero retention” standard, your data is never stored outside of Salesforce. For more information, check out the Einstein Trust Layer badge. 
   
   
   
5. Testing and refinement-Prompt Builders typically provide a workspace to test prompt templates in real-time. Users can preview how the AI will respond to specific data, allowing them to iterate and refine the prompt until it consistently delivers the desired results.

 

Different types of Prompt Templates-

• Sales Email prompt templates help your team draft truly personalized emails for your customers, products, and events based on record data.
• Field Generation prompt templates bring generative AI-assisted workflows to custom fields within a Salesforce record.
• Record Summary prompt templates create a rich-text summary for a Salesforce record based on the record’s data.
• Flex prompt templates generate content for business purposes not covered by other prompt template types, allowing you to define your own resources. Flex prompt templates in Prompt Builder offer a versatile and efficient way to automate the creation of text for PDFs, social media posts, newsletters, emails, record fields, and more with Einstein generative AI and the LLM (large language model). Unlike the Sales Email or Field Generation prompt templates in Prompt Builder, Flex templates don’t have predefined entry points. What does this mean? You can select any object in your organization to use in the template and apply the completed prompt wherever you see fit. Whether you're using Apex, Flow, REST API, or Agentforce, Flex templates are designed to fit your needs.

What are the Permission Sets involved?

Prompt Template User

Role: End user or consumer of prompts
What they can do:
View published prompt templates
Use templates in supported Salesforce interfaces (e.g. console, agents, flows)
Input values and run prompts in supported channels (like Knowledge, Case replies, or custom use cases)

2.      Prompt Template Manager

Role: Builder, administrator, or prompt engineer
What they can do (in addition to user permissions):
Create new prompt templates
Edit and delete templates
Test templates in the Prompt Builder UI
Manage prompt versions (draft, published, etc.)
Set input variables, retrievers, responses, resolution logic, etc.

 

 

 

 

 

 

DevOps Center​

• DevOps Center lets you choose whether to manage your releases from its point-and-click interface, or directly from the source control system, or a combination of both. Under the hood, we manage the source control branches so developers and builders can focus on development tasks.​
• Because all changes are captured in a source control system, you have a single source of truth for configuration and code, which improves collaboration across all functions: admins, developers, release managers, QA, and other business stakeholders.​

What do you need to start with DevOps Center?

• Access to DevOps Center in Salesforce​
• Access to Dev Org, Dev Integrated Org, QA Org, and Stage Org​
• Access to GitHub, (BB is in Beta as of 11/26/2024)

Work Item-

In Salesforce DevOps Center, a work item is a logical unit of work that represents a collection of metadata changes for a project​

Because all changes are captured in a source control system, you have a single source of truth for configuration and code, which improves collaboration across all functions: admins, developers, release managers, QA, and other business stakeholders.​

A work item encapsulates information about an objective or task while it’s being worked on and is assigned to one individual. The work item is a historical record if questions arise about those changes. ​

Create Review ​

• Once work item is created and components get added, pls submit for review by clicking the “Create Review” button on Work item.​
• When a work item is reviewed and approved, you can change the status to Ready to Promote.​

Recommendation on Work Item​

• If a plan changes, or if a change conflicts with another change, you can change the status of a work item to Never. This will make the work item inactive and return the committed files to the list of available changes. ​
• Limit the number of work items​
• Combine work items​

Things to Remember​

• SF Devops center is in its initial stages, It's not as rich as other tools like Flosum or Copado or many more.​
• We will not have the cool feature that flosum had which is the Overwrite protection against the org.​
• BitBucket is support is still in Beta version