AI & Intelligence

Servantium integrates AI throughout the engagement lifecycle — not as a standalone feature, but as an intelligence layer that makes every part of the system smarter over time. This guide covers all AI capabilities and how they work together.

Overview

Servantium’s AI is powered by Google Vertex AI (Gemini models) and Firestore Vector Search. The AI operates primarily through backend Cloud Functions that trigger automatically, keeping the intelligence layer invisible until you need it.

Capability	AI model	Trigger
Similar engagement matching	Gemini Embedding 001 + Vector Search	Automatic (on note save)
AI quote generation	Gemini 2.5 Flash	Manual (user clicks generate)
Note context extraction	Gemini 2.5 Flash	Manual (user clicks AI button)
Engagement template generation	Gemini	Manual (admin provides instructions)
Service Catalog learning	Rule-based (no model)	Automatic (on quote save)

Similar engagement matching

The Similars feature finds past engagements that resemble the current one, using semantic understanding rather than keyword matching.

How it works

1. Embedding generation — When notes are saved on an engagement, a backend trigger (noteEmbedding) aggregates all notes, generates a vector embedding using Gemini Embedding 001, and stores it on the engagement document.

2. Similarity search — When you open the Similars tab, a Cloud Function (getsimilar) performs a vector search across all engagement embeddings in your organization, returning the most semantically similar past engagements.

3. Results display — Similar engagements appear as summary cards showing the engagement name, account, and key details. Use these to reference pricing, scope, and delivery outcomes from comparable work.

Why it matters

Traditional search requires you to remember exact keywords or project names. Semantic matching understands context:

• A “CRM modernization” engagement matches against a past “customer platform migration”
• A “data warehouse consolidation” finds similar “analytics infrastructure” projects
• The matching improves automatically as more engagements are completed

AI quote generation

Servantium can generate a complete quote — with sections, line items, quantities, costs, and markups — based on your engagement’s context.

How it works

1. Trigger — Click the AI generate button (sparkle icon) on a quote. This sets ai_generate = true on the quote document.

2. Backend processing — The quoteAIGenFunction Cloud Function reads:

   • All notes on the engagement
   • Data from similar historical engagements (found via vector search)
   • The organization’s Service Catalog (master item list with costs and markups)

3. Generation — Gemini 2.5 Flash generates a detailed quote structure with appropriate sections, items, quantities, and pricing based on the engagement’s context and historical patterns.

4. Result — The generated quote appears in the editor. Review, adjust, and refine as needed.

Prerequisites

AI quote generation works best when:

• The engagement has notes that describe the client’s needs
• The engagement has a template assigned (provides structural context)
• Your organization has historical engagements with notes and completed quotes
• The Service Catalog has items with current costs and markups

Note context extraction

When an engagement has both notes and a template assigned, AI can automatically extract structured data from your notes and populate the template fields.

How it works

1. Trigger — Click the AI button (sparkle icon) next to the Type dropdown on the engagement’s Info tab.

2. Processing — Servantium reads the engagement template’s field definitions and all notes, then uses Gemini 2.5 Flash to:

   • Parse the unstructured note text
   • Identify information relevant to each template field
   • Return structured JSON that maps to the template’s schema

3. Result — The engagement’s custom data fields are populated with the extracted values. Text fields get text, chip selectors get appropriate options, sliders get numeric values.

Example

If your template has fields for “Client Industry,” “Project Timeline,” “Budget Range,” and “Key Technologies,” and your notes mention that the client is a healthcare company looking to implement a new EHR system within 6 months with a budget of $500K-$750K, the AI extracts and maps each piece of information to the correct field.

Engagement template generation

Organization admins can use AI to generate or modify engagement templates from natural language descriptions.

The generate_engagement_template Cloud Function accepts natural language instructions (e.g., “Create a template for Salesforce implementations with fields for org size, number of users, current CRM, timeline, and budget”) and returns a configured template with appropriate field types, labels, and options.

The Learning Catalog

The Service Catalog is not just a static reference — it is a learning system that improves with use.

How it works

The quoteCatalogFunc backend trigger fires whenever a quote is modified. It:

1. Scans the quote for new items or items with updated costs/markups
2. Checks each item against the master Service Catalog
3. New items are added to the catalog automatically
4. Updated costs/markups on existing items refresh the catalog values

Over time, this creates a comprehensive, up-to-date catalog that reflects your organization’s actual pricing — not just what someone set up initially.

Impact on AI quote generation

The Learning Catalog directly feeds AI quote generation. When the AI generates a quote, it consults the catalog for accurate, current pricing. As the catalog grows and stays current, AI-generated quotes become more accurate.

Document versioning

While not strictly AI, Servantium’s versioning system supports the intelligence layer by maintaining a complete history of changes.

The handle_versioning_event function archives the previous state of documents (quotes, engagements, etc.) into a versions collection whenever they are modified. This provides:

• Full audit trail of changes
• Ability to reference previous versions
• Historical data for pattern analysis

The institutional memory engine

All of these AI capabilities work together as Servantium’s institutional memory engine:

Notes --> Embeddings --> Similar Matching --> AI Quote Generation
  |                                              |
  +-- Context Extraction                         |
                                                 v
                              Learning Catalog <-- Quote Modifications

The cycle:

1. Capture — Team members record notes during client interactions
2. Structure — AI extracts structured data from notes into template fields
3. Embed — Backend triggers generate vector embeddings from notes
4. Match — New engagements find similar past work through vector search
5. Generate — AI builds quotes using notes, similar engagements, and the Service Catalog
6. Learn — Quote items flow back into the Service Catalog; completed engagements strengthen future matching
7. Repeat — Each completed engagement makes the next estimate more accurate

AI principles

Servantium’s AI follows three principles:

1. Assist, don’t decide. Every AI output is a suggestion that can be reviewed and overridden. Human judgment stays in control.
2. Show your work. Similar engagements show you the actual past work the AI is drawing from. You can evaluate the reasoning, not just the result.
3. Improve with use. The system gets better over time through the Learning Catalog and growing engagement history — no manual training required.

What’s next?

• Notes & AI — Deep dive into notes and how they feed AI
• Quotes & Pricing — AI quote generation in practice
• Engagements — Similar matching and context extraction on engagements
• Settings & Administration — Manage the Service Catalog and templates