PitchCraft

Turn discovery calls into winning proposals automatically. Quote prediction using Machine Learning as a Tool and proposal generation for agencies who want to close faster without the manual work

Comment

Inspiration

As an AI automation agency owner, we spent countless hours manually crafting proposals after discovery calls—analyzing transcripts, researching prospects, and calculating project quotes. I saw agencies losing deals due to slow proposal turnaround and inconsistent pricing. PitchCraft was born from the need to automate this entire workflow while maintaining the personalized touch that wins clients.

What it does

PitchCraft transforms discovery call recordings into professional, customized proposals with accurate pricing predictions. It uses a research agent with tool calls to analyze call transcripts and gather prospect intelligence, then leverages a pre-trained XGBoost ML model via FastAPI to predict project quotes in real-time based on integration complexity, scope, and technical requirements. with Machine Learning as a Tool (MLAT) as a nuianced framework that modern literature rarely uses.

How we built it

Built on n8n workflow automation with a multi-agent architecture. The research agent uses tool calls to analyze transcripts and research prospects. The ML pricing engine runs XGBoost through FastAPI, predicting quotes based on complexity factors. All workflows orchestrate seamlessly to generate complete proposals automatically.

Challenges we ran into

Integrating real-time ML predictions as tool calls within n8n flows required custom API architecture. Balancing automation speed with research depth, and ensuring the XGBoost model accurately predicted pricing across diverse project types with VERY limited training data. (we only have 40 real data and 30 are synethetic by LLM!)

Accomplishments that we're proud of

Created a fully functional agentic system that reduces proposal creation from hours to minutes. Successfully integrated ML-powered pricing predictions as tool calls within the agent framework, delivering accurate quotes (sort of) I actually use this now everyday for my agency!

What we learned

How to architect multi-agent systems for complex business workflows. The power of combining research agents with ML prediction models through tool calls. Real-world ML deployment challenges and the importance of feature engineering for pricing models.

What's next for PitchCraft

Expanding the ML model with more training data for improved accuracy. Adding CRM integrations for automated proposal delivery. Building a feedback loop to continuously improve pricing predictions based on won/lost deals. Creating industry-specific proposal templates and pricing models.

More importantly - extend this to ANY quote based businesses in real time, like Construction!

Built With

Share this project:

Updates

posted an update —

Market Opportunity

The Quotation Bottleneck

Every quote-based business faces the same fundamental challenge: the faster you quote, the more likely you win. Yet proposal generation remains one of the most time-intensive, manually-driven processes across entire industries.

Industries plagued by slow quotation cycles:

  • Construction & Contracting: Material costs, labor estimates, site assessments—quotes can take days or weeks
  • Professional Services: Consulting, legal, accounting firms spend hours scoping and pricing custom engagements
  • Home Services: HVAC, roofing, plumbing, landscaping—technicians often quote on-site or delay proposals
  • Manufacturing & Custom Fabrication: Complex BOMs, tooling costs, and lead times require extensive calculations
  • Event Planning & Catering: Venue-dependent pricing, guest counts, menu customization
  • Marketing & Creative Agencies: Scope creep, unclear requirements, subjective deliverables
  • IT & Managed Services: Infrastructure assessments, licensing, support tiers
  • Insurance & Financial Services: Risk assessment, underwriting, compliance considerations

The Universal Problem

Across all these industries, the pattern is identical:

  1. Discovery conversation happens (call, site visit, consultation)
  2. Manual research required (client background, technical requirements, competitive landscape)
  3. Pricing calculations performed (often inconsistent, experience-based, prone to error)
  4. Proposal drafting takes hours or days
  5. Client receives quote after momentum has cooled

Speed-to-lead is everything. Studies show that responding within 5 minutes vs. 30 minutes decreases conversion by 21x. Yet most quote-based businesses take days.

PitchCraft's Addressable Market

Total Addressable Market (TAM):

  • Quote-based B2B services represent a $2+ trillion market in the US alone
  • Construction industry: $1.8T annually
  • Professional services: $500B+ annually
  • Home services: $600B+ annually

Serviceable Addressable Market (SAM):

  • Businesses with 5-100 employees who quote custom projects
  • Estimated 5M+ businesses in North America alone
  • Average proposal cost: $500-2,000 in labor per quote

Serviceable Obtainable Market (SOM):

  • Early adopters: Tech-forward SMBs in consulting, agencies, and specialty contractors
  • Estimated 50K businesses ready to adopt AI quotation systems now
  • At $200-500/month per business = $120M-300M annual opportunity

Why Now?

Three converging factors make this the ideal moment:

  1. LLM tool-calling maturity: Gemini 3 and similar models can reliably orchestrate multi-step workflows
  2. ML democratization: XGBoost and other frameworks enable accurate predictions with limited data
  3. Automation infrastructure: n8n, Zapier, Make.com have normalized workflow automation

The Vision: Universal Quote Automation

PitchCraft's architecture is industry-agnostic. The same system that quotes AI automation projects can quote:

  • Construction projects: Material costs, labor hours, permit fees
  • Consulting engagements: Hourly rates, deliverable scope, timeline
  • Home service jobs: Parts, labor, travel, warranty
  • Manufacturing runs: Materials, tooling, setup, per-unit costs

The core components remain constant:

  • Conversation capture (call recording, site notes, consultation forms)
  • Research agent (company intel, technical requirements, constraints)
  • ML pricing model (trained on historical quotes and outcomes)
  • Proposal generation (professional documents with accurate pricing)

Only the feature engineering and training data change per industry.

Competitive Moat

Most quotation software is:

  • Static templates (Proposify, PandaDoc) without intelligence
  • Industry-specific (construction estimating, legal billing) without ML
  • Enterprise-focused (Salesforce CPQ) too complex for SMBs

PitchCraft combines:

  • AI research agents for personalization
  • ML pricing predictions for accuracy
  • Full automation for speed
  • SMB-friendly deployment (no-code workflows)

Impact at Scale

If PitchCraft achieves even 1% penetration of the SAM (50,000 businesses):

  • 500 businesses automated = $60M-150M saved annually in proposal labor
  • 2-3 day turnaround → 2-4 hour turnaround = 10x faster speed-to-lead
  • Inconsistent pricing → ML-standardized pricing = higher win rates, better margins

The real transformation is democratizing enterprise-grade sales intelligence for SMBs, enabling small businesses to compete on speed and professionalism with much larger competitors.

PitchCraft isn't just a tool for agencies. It's the blueprint for automating quotation across every project-based business in the economy.

Log in or sign up for Devpost to join the conversation.

posted an update —

The Agents Part

PitchCraft operates through two primary AI workflows that orchestrate the entire proposal generation pipeline, leveraging Gemini 3's tool-calling and structured output parsing capabilities for intelligent research and XGBoost for pricing predictions.

Architecture Overview

The system uses a multi-agent architecture where specialized AI agents handle distinct phases of the proposal workflow:

┌─────────────────────────────────────────────────────────────────┐
│                     WORKFLOW 1: Research Agent                   │
├─────────────────────────────────────────────────────────────────┤
│  Fireflies → Transcript Extract → Gemini Research Agent →       │
│  → [Tool Calls: Company Intel, Tech Stack, Competitor Analysis] │
│  → Structured Output → Draft Agent                              │
└─────────────────────────────────────────────────────────────────┘
                              ↓
┌─────────────────────────────────────────────────────────────────┐
│                  WORKFLOW 2: Pricing & Generation                │
├─────────────────────────────────────────────────────────────────┤
│  Draft Input → Feature Extraction → XGBoost FastAPI Call →      │
│  → Price Prediction → Gemini Proposal Assembly →                │
│  → Document Generation → Client Delivery                        │
└─────────────────────────────────────────────────────────────────┘

Workflow 1: Research & Intelligence Agent

Research Agent Workflow

Agent Flow:

  1. Webhook Trigger → Receives Fireflies transcript data
  2. Code in JavaScript → Extracts and structures transcript content
  3. Get Path Score → Retrieves conversation metadata
  4. Information Extractor → Gemini 3 agent extracts key requirements
  5. Research Agent (Multi-Tool) → Gemini 3 with tool calls:
    • Firecrawl API → Look up client annual revnue
    • Perplexity API → Company + Prospect background research
    • Simple Memory → Stores research context
    • Batch Processing → Parallel tool execution
    • Project Lookup → Internal database queries
    • Prep Predictions → Prepares data for ML model
  6. Google Gemini 3 Agent → Synthesizes research findings
  7. Draft Agent → Structures initial proposal outline
  8. JavaScript → Formats output for next workflow
  9. Update Document → Stores research in Google Drive
  10. Respond to Webhook → Passes data to pricing workflow

Key Innovation: Gemini 3 orchestrates multiple tool calls in parallel, gathering company intelligence, competitive analysis, and technical requirements simultaneously—mimicking how a human sales engineer would research before quoting.

Workflow 2: Pricing Prediction & Proposal Generation

Pricing & Generation Workflow

Agent Flow:

  1. Webhook Trigger → Receives research output from Workflow 1
  2. Information Extractor → Gemini 3 extracts pricing features
  3. Get Path Score → Retrieves complexity metrics
  4. Integration Gauge (Research Model) → Scores integration complexity
  5. Edit Fields → Structures data for ML model
  6. Research Agent → Gemini 3 with ML tool call:
    • Google Gemini Chat Model → Analyzes pricing context
    • Structured Output (Parser) → JSON validation
    • Simple Memory → Maintains conversation state
    • Prep Predictions → Formats features for XGBoost
    • Google Gemini Chat Model (Pricing) → Contextualizes predictions
  7. Draft Agent → Assembles final proposal with ML pricing
  8. Code in JavaScript → Formats proposal document
  9. Copy File → Creates proposal template
  10. Update a Document → Generates final Google Doc by having a template with placeholder text. Using the JSON fields we can then use the Find & Replace Text function in Google Docs API
  11. Download File → Prepares deliverable
  12. Respond to Webhook → Delivers proposal to client system

Key Innovation: The XGBoost FastAPI endpoint is called as a tool within the Gemini 3 agent workflow. The agent doesn't just receive a price—it reasons about the prediction, explains the pricing rationale, and integrates it into a coherent narrative.

Tool-Calling Architecture

Both workflows leverage Gemini 3's native tool-calling capabilities to create an agentic system where:

  • Research tools (Google Search, database queries) run in parallel
  • ML predictions (XGBoost via FastAPI) execute as agent tools
  • Memory systems maintain context across tool calls
  • Structured outputs ensure reliable JSON responses for downstream processing

This "ML as a Tool" pattern is rare in production systems—most implementations separate LLM reasoning from ML predictions. PitchCraft treats the XGBoost model as just another tool the agent can call, enabling it to reason about pricing in context rather than blindly inserting numbers.

Workflow Orchestration

The n8n platform enables:

  • Asynchronous execution → Research and pricing happen in parallel where possible
  • Error handling → Fallbacks for failed tool calls or API timeouts
  • State management → Memory nodes maintain context across agent calls
  • Webhook chaining → Workflows trigger each other seamlessly
  • Real-time monitoring → Visual debugging of agent decision paths

Total execution time: 2-3 minutes from Fireflies transcript → delivered proposal, compared to 3+ hours manually.

Log in or sign up for Devpost to join the conversation.

posted an update —

The ML Part

Training Strategy:

With only 70 total examples (40 real deals + 30 LLM-generated synthetic data), we used XGBoost regression with 3-fold cross-validation to maximize predictive power while preventing overfitting.

Feature Engineering:

  • Integration complexity scoring
  • Scope quantification (feature count, custom logic requirements)
  • Technical stack requirements (platform complexity, data volume)
  • Client segment indicators (industry, company size)
  • Client Pain Severity (How painful + urgent are they?)

Model Performance:

Metric Training Set Test Set Cross-Validation
R² 0.937 0.807 0.816 ± 0.060
MAE $2,328 $3,688 $3,898 ± $629
RMSE $2,874 $4,720 -

Key Results:

  • 80.7% explained variance on unseen data
  • Predictions within ±$3,688 on average, commercially viable for real quotes
  • CV-Test consistency: Test R² (0.807) aligns with CV estimate (0.816), confirming reliable generalization
  • Stable across folds: R² ranged from 0.74-0.88, showing model robustness

Deployment:

  • Serialized via joblib, served through FastAPI
  • Real-time inference with <100ms latency
  • Integrated as a tool call within the Gemini 3 agent workflow
  • Returns predictions with confidence intervals for quote ranges

The model proves that even with limited training data, aggressive feature engineering and careful cross-validation can produce production-ready ML systems for high-stakes business decisions.

Log in or sign up for Devpost to join the conversation.

posted an update —

A little bit of Extra Write up on The Tech Side:

Built with n8n + Gemini 3 API + XGBoost

Tech Stack:

  • n8n – Multi-agent workflow orchestration
  • Gemini 3 API – Research agent and proposal draft writing agent with tool-calling capabilities as well as parsing complicated structured output consistently and accurately
  • Fireflies.ai – Call transcription logs
  • XGBoost – ML pricing prediction (3-fold CV)
  • FastAPI – Real-time ML inference endpoint

System Architecture

Multi-Agent Pipeline

Fireflies Call → Gemini 3 Research and Transcript Analysis Agent (Tool Calls) → Gemini 3 Proposal Draft Agent with
XGBoost Pricing (as a FastAPI) → Proposal Generation

1. Intelligent Research + Analysis Agent (Gemini 3)

  • Analyzes discovery call transcripts in detail
  • Uses tool calls to research prospects (company intel, tech stack, competitors)
  • Extracts requirements and complexity factors (e.g annual client revenue, integration complexity, tech stack, pain points of clients)
  • Gathers context for personalized proposals
  • Parse a complex JSON as output for proposal writing

2. Proposal Drafting Agent (Gemini 3 + ML Pricing Engine (XGBoost + FastAPI))

  • Predicts project quotes based on:
    • Integration complexity
    • Project scope
    • Technical requirements
  • Trained on 40 real deals + 30 LLM-synthetic examples
  • Deployed as FastAPI endpoint for tool-call integration
  • Then draft various sections of the proposals and parse different sections of proposals as a JSON field.

3. Machine Learning as a Tool (MLAT)

  • Novel framework treating ML models as agent tools
  • Real-time predictions within agentic workflows
  • Rarely explored in modern AI literature

4. Real Time Discovery Agent (Gemini 3 Flash)

  • Another mode of PitchCraft where the client themselves can talk to the discovery agent, and based on that transcript, it will do the above analysis

What Makes This Unique

ML with Limited Data: Successfully deployed XGBoost with only 70 total examples (40 real + 30 synthetic) through aggressive feature engineering (to 6) and CV tuning.

Gemini 3 Tool Integration: Research agent dynamically calls ML pricing API as a tool, combining LLM reasoning with statistical prediction.

Real Business Impact: We actually use it daily in our agency sales process:

Speed-to-Lead: Proposals delivered within 2-4 hours of discovery call vs. 2-3 days previously

Time Savings: Reduced proposal creation from 3+ hours to under 10 minutes of hands-on work

Consistency: Eliminated pricing variance—every quote now uses the same ML-driven methodology

Log in or sign up for Devpost to join the conversation.