A collection of intelligent agent workflows built with Google's Agent Development Kit (ADK) to enhance email campaign targeting and outreach automation for Fiona.
Fiona is a platform designed to simplify email campaign management and communications. It enables users to create, manage, and track email campaigns with clear analytics and contact organization—all without requiring technical expertise. Learn more at fiona.mrphilip.cv.
These agents extend Fiona's capabilities by automating intelligent research, problem discovery, and personalized outreach generation. They power the next generation of Fiona's targeting and campaign automation features.
File: career-outreach-sequential-agents.ipynb
This workflow combines three specialized agents in a strict sequence:
- Problem Finder Agent - Identifies industry-specific challenges
- Problem Solver Agent - Maps how your career skills address those problems
- Cold Outreach Message Agent - Generates personalized LinkedIn messages
How it fits Fiona: When users want to launch targeted cold email campaigns, this agent automatically discovers relevant problems their audience faces and generates customized email templates. Users simply input their career and target industry, and the agent creates ready-to-send message templates directly compatible with Fiona's campaign creation system.
Integration: Output can be imported into Fiona as email templates or bulk contact lists with pre-generated subject lines.
File: parallel-cold-email-outreach.ipynb
This workflow runs three research agents simultaneously, then synthesizes findings:
- Reddit Research Agent - Gathers cold outreach tactics from discussions
- Quora Research Agent - Collects expert best practices
- Freelance Niche Agent - Identifies service opportunities
- Email Aggregator Agent - Creates 3-4 tailored email templates
How it fits Fiona: Speed is critical in email campaigns. By running research in parallel, this agent dramatically reduces analysis time. Users provide their career field once, and within seconds receive multiple email templates ready for use. This accelerates Fiona's campaign creation workflow.
Integration: Generated email templates integrate directly into Fiona's template library. The parallel execution model allows Fiona to offer near-instant email suggestions without user wait times.
File: personalized-lead-outreach.ipynb
This notebook demonstrates the Agent Tools pattern where agents are used as tools within other agents:
- Researcher Agent - Conducts comprehensive lead research using Google Search
- Main Outreach Agent - Uses researcher as a tool to gather intelligence, then generates personalized emails
Features:
- Multi-step research gathering (role, company, background, social presence, achievements)
- Research-backed email personalization
- Clear separation of concerns (specialist agents)
- Reusable researcher agent for other workflows
- Simple two-variable input (career field + lead name)
How it fits Fiona: This demonstrates how agents can specialize and delegate work. Users input a lead name and career field, the researcher agent gathers intelligence, and the main agent produces highly personalized outreach emails ready for Fiona campaigns.
Integration: Generated emails can be imported directly into Fiona's template library or used as starting points for bulk campaigns.
File: loop-agent-problem-refinement.ipynb
This workflow discovers and refines problems through iterative passes:
- Problem Finder Agent - Discovers 10-15 initial problems
- Critic Agent (Loop) - Iteratively eliminates weak/non-existent problems (up to 3 passes)
- Final Output Agent - Formats each problem as a one-sentence statement
How it fits Fiona: Quality of email lists matters. This agent validates that discovered problems are real and substantial, ensuring Fiona users build campaigns targeting genuine, location-specific opportunities. The loop refinement ensures data accuracy across multiple iterations.
Integration: Refined problem lists become targeting criteria in Fiona. Users can segment contact lists based on validated industry problems, improving campaign relevance and conversion rates.
File: human_in_loop_lead_outreach.py (Standalone Script)
This production-ready workflow implements intelligent lead research with human approval and AI regeneration:
- Researcher Agent - Conducts comprehensive lead research using Google Search
- Outreach Agent - Generates personalized cold outreach emails
- Approval Agent - Submits emails for human review (pausable workflow)
- Delivery Agent - Sends approved emails or triggers regeneration
Features:
- Personalized lead research (company, role, achievements, social presence)
- AI-generated customized emails (subject + body)
- Human-in-the-loop approval workflow (approve/reject/edit/regenerate)
- Automatic regeneration on rejection (up to 3 attempts)
- Mock email sending (ready for Fiona/SendGrid integration)
- Interactive command-line interface for approvals
How it fits Fiona: This script demonstrates production-quality workflows that maintain human oversight. Users can research leads, approve generated emails, and send through Fiona with confidence that each message has been reviewed. The regeneration loop ensures email quality meets user standards before sending.
Integration:
- Output emails can be imported directly into Fiona campaigns
- Approval workflow integrates with Fiona's UI for user decisions
- Mock send can connect to Fiona backend or SendGrid
- Session state persists across long approval cycles
Usage:
python human_in_loop_lead_outreach.pyThen interact with the approval menu to review, edit, or regenerate emails before sending.
File: lead_memory_research.py (Standalone Script)
This production script demonstrates persistent session management with SQLite database:
- Researcher Agent - Gathers comprehensive lead information using Google Search
- Memory Agent - Asks multiple questions about the lead, remembering information across turns
- Session State - Stores research in
lead:nameandlead:researchkeys - Database Persistence - Uses SQLite with async support (aiosqlite)
- Interactive Terminal - Continuous conversation with lead memory
Features:
- 🧠 Research a lead once, ask unlimited questions about them
- 💾 All data persists in SQLite database (survives application restarts)
- 📝 Session state automatically tracks lead information
- 🔍
databasecommand to inspect what's stored in SQLite infocommand to view current lead data in memory- Multi-session support (different leads in different sessions)
How it fits Fiona: This demonstrates how to build agents that maintain long-term lead intelligence. Sales teams can research prospects once, then ask contextual questions across multiple sessions without re-researching. All lead data persists in the database, enabling better targeting and follow-up campaigns.
Integration:
- Lead research and memory data flows to Fiona's CRM system
- Session state can be exported to Fiona contact profiles
- Enables lead scoring based on research findings
- Supports multi-turn lead qualification workflows
Usage:
python lead_memory_research.pyExample Workflow:
You > research Satya Nadella
(Agent researches and saves to database)
You > What companies has he worked at?
(Agent retrieves saved research, answers from memory)
You > What are his recent achievements?
(Same research, new question - all from database)
You > database
(Inspect SQLite database structure and contents)
You > quit
(Exit - lead research persists in database!)
Key Technical Features:
DatabaseSessionServicewithsqlite+aiosqlite://for async SQLite- Auto-created database file:
lead_research_memory.db - Session state management with
tool_context.state - Resumable apps that survive interruptions
These agents demonstrate three core ADK patterns that enhance Fiona:
Used when order matters and each step builds on the previous one.
- Problem discovery → Skill mapping → Message generation
Used when independent tasks can run simultaneously for speed.
- Reddit research + Quora research + Freelance research (all at once)
Used when agents need to specialize and delegate work to other agents.
- Researcher agent (finds data) + Main agent (uses research to craft output)
Used for iterative refinement and quality assurance.
- Discover problems → Review & eliminate weak ones (repeat) → Format output
| Feature | Agent | Benefit |
|---|---|---|
| Smart Campaign Targeting | Loop Agent (Problem Refinement) | Validated, location-specific targeting criteria |
| Template Generation | Sequential + Parallel Agents | Auto-generated email templates save hours |
| Multi-Source Research | Parallel Agent | Fast, comprehensive market research in seconds |
| Career-Based Segmentation | Sequential Agent | Target users by career field and industry skills |
| Rapid Prototyping | All Agents | Test campaign angles instantly before sending via Fiona |
| Agent Specialization | Personalized Lead Outreach (Agent Tools) | Researchers and outreach specialists work independently |
| Research-Backed Personalization | Personalized Lead Outreach (Agent Tools) | Authentic emails with genuine lead insights |
| Human-Approved Outreach | Human-in-the-Loop Script | AI generates, humans approve, regenerate until satisfied |
| Lead Intelligence | Human-in-the-Loop Script | Research-backed personalization increases engagement |
| Multi-Turn Memory | Lead Memory Research (Persistent Sessions) | Ask unlimited questions about leads, all remembered |
| Database Persistence | Lead Memory Research (Persistent Sessions) | Lead research survives restarts, enables long-term tracking |
| Lead Qualification | Lead Memory Research (Persistent Sessions) | Interactive questioning builds complete prospect profiles |
- Python 3.8+
- Google API key (Gemini API)
.envfile withGOOGLE_API_KEY
- Clone or download this folder
- Create a
.envfile in the mira directory:GOOGLE_API_KEY=your_api_key_here - Install dependencies:
pip install -r requirements.txt
- Open any notebook in Jupyter or VS Code
- Modify the input variables (career, industry, city as needed)
- Run the cells to generate insights
- Copy outputs directly into Fiona campaigns
- Input: Career background, target industry
- Output: Personalized LinkedIn messages for each opportunity
- Input: Career field
- Output: 3-4 ready-to-send email templates with subject lines
- Input: Career field, lead name
- Output: Lead research summary, personalized cold outreach email
- Input: Career, industry, city
- Output: Validated list of problems (one sentence each)
- Input: Career field, lead name, lead email
- Output: Researched lead profile, personalized email (subject + body), user approval status, sent message ID
- Input: Lead name (first query), unlimited follow-up questions
- Output: Research findings, answers to questions based on remembered data, session state persisted in SQLite
These agents are designed to integrate with Fiona's Django backend:
- Campaign Template Import - Agents generate templates that import into Fiona's template library
- Contact List Segmentation - Validated problems become targeting filters
- Bulk Email Generation - Multi-template output enables A/B testing in Fiona
- Analytics Integration - Track which agent-generated templates perform best
- Lead Memory & Persistence - Lead Memory Research script can:
- Export session state to Fiona contact profiles
- Store lead research data in Fiona CRM
- Enable multi-turn lead qualification workflows
- Build complete prospect profiles across sessions
- Support long-term lead tracking and scoring
Architecture Example - Lead Memory Integration:
Fiona Frontend
↓
[User enters: Lead name]
↓
Lead Memory Research Agent
├─ Research (one-time, stored in SQLite)
└─ Questions (unlimited, always using stored research)
↓
Session State (persisted in lead_research_memory.db)
├─ lead:name
├─ lead:research
└─ Additional custom fields
↓
Export to Fiona
├─ Lead research → Contact profile
├─ Session history → Interaction log
└─ Memory state → CRM fields
↓
Multi-session Lead Tracking
└─ Can resume where you left off, even after restart
mira/
├── career-outreach-sequential-agents.ipynb # Sequential workflow (notebook)
├── parallel-cold-email-outreach.ipynb # Parallel workflow (notebook)
├── personalized-lead-outreach.ipynb # Agent Tools pattern (notebook)
├── loop-agent-problem-refinement.ipynb # Loop workflow (notebook)
├── human_in_loop_lead_outreach.py # Human approval + regeneration (standalone script)
├── lead_memory_research.py # Persistent sessions + memory (standalone script)
├── lead_research_memory.db # SQLite database (auto-created)
├── requirements.txt # Dependencies
├── .env # API keys (gitignored)
├── .gitignore # Excludes sensitive files
└── README.md # This file
The .gitignore file excludes:
.env(contains API keys)day-1a-from-prompt-to-action.ipynb(learning notebook)day-1b-agent-architectures.ipynb(learning notebook)
Only production agents are tracked.
Planned enhancements to deepen Fiona integration:
- Real-time agent suggestions as users create campaigns
- Agent-powered contact list enrichment
- Automated A/B testing of agent-generated templates
- Performance scoring for agent outputs
- Multi-language agent support
These agents are part of the Fiona ecosystem. Use them to enhance your Fiona campaigns.
Questions or feedback? Reach out or visit fiona.mrphilip.cv to learn more about how agents power intelligent email campaigns.