🌍 LLM Agent Lab

AI agent lab built with LangChain + LangGraph.
It’s a sandbox where ideas turn into working MVPs: small, functional agents that can later be exposed as APIs and consumed in Next.js/React projects to create dynamic and scalable websites.

The vision: build a library of agents, each one tackling a real-world use case, starting lean and evolving into production-ready tools.

---

🚀 Current MVP Agents (Catalog)

All current MVP agents live under the /examples/ folder as Jupyter notebooks or Gradio demos.

1. 📰 Grow Pulse Daily Reader (examples/grow-pulse-daily-reader.ipynb)

• Generates daily readings with 3–5 recent news items related to the user’s profession and sector.
• Extracts opportunities contextualized for the career path.
• Suggests one daily actionable step (≤15 min).
• Drafts bilingual LinkedIn posts (Spanish + English).
• Creates 3 visible mini-POCs (≤45 min).
• Explains how actions, posts, and POCs compound into career growth.

2. 🤖 AI Daily Reader / Career Coach (examples/ai_daily_reader.ipynb)

• Focuses specifically on AI industry news (OpenAI, Anthropic, DeepMind, OSS, enterprise adoption).
• Maps news items into opportunities for Tech Leads and Full-Stack profiles.
• Suggests one micro-action to get closer to global consulting opportunities (+10K/month).
• Drafts bilingual LinkedIn posts (authoritative, inspiring, non-egocentric).
• Provides POC ideas tied to AI and .NET scenarios.
• Explains how daily actions + posts + POCs compound strategically.

3. 🛠️ Weekend Support Agent (examples/pura_vida_helper.ipynb)

• Connects to Confluence space as a weekend support helper.
• Retrieves docs (via embeddings + FAISS).
• Summarizes docs in plain language.
• Suggests one practical action (≤15 min).
• Provides a stitched final support answer.
• ✅ Supports Prompt Pages (special Confluence pages that act as agent instructions).
  e.g. AI-Agent: Service Desk Instructions, AI-Agent: Travel Experience Playbook

4. 🧠 Grow Routine – Nutrition & Lifestyle Validator (examples/macro_nutrition_pipeline.ipynb)

• Spanish-first multi-role agent combining:
  • 🏋️ Coach → Generates physical & lifestyle recommendations.
  • 🥦 Nutritionist → Creates nutrition plan with macros, calories, considerations (fasting, allergies), and technical justification.
  • 🔎 Macro Validator → Validates consistency of macros, calories, and reasoning.
• Conditional loop: nutrition plan regenerates automatically if validation fails.
• Outputs a final Markdown summary: plan + validation + motivational note.
• ✅ Interactive Gradio UI for InBody + goals input → final structured plan output.

5. ✉️ GrowRoutine Habit → Email Generator (examples/habit_to_email.ipynb)

• Turns Pareto 80/20 habits into practical, motivational email campaigns.
• Workflow:
  • 🧠 TrendScout → Extracts 3–5 core habits (name, why it’s 80/20, minimal action, principle, example).
  • ✍️ Marketing Agent → Picks one habit and writes a structured email (subject, preheader, hook, steps, why it works, book reference, weekly challenge, CTAs, signature).
  • ✅ Validator → Ensures one-habit focus, correct structure, clarity, length (140–220 words), and energy.
• Conditional feedback loop: regenerates emails if validation fails.
• Outputs a final Markdown deliverable (habits + email + validation).
• Includes an interactive Gradio app where users can enter campaign briefs and instantly generate approved habit-based emails.

6. 🍲 Grow Routine – Cultural Meal Plans (examples/cultural_meal_plans.ipynb)

• Spanish-first multi-role agent combining:
  • 🏋️ Coach → Generates personalized physical & lifestyle recommendations.
  • 🥦 Nutritionist → Creates 3 full daily menus (Day 1, 2, 3) with:
    • Lunch, post-workout snack, dinner.
    • Ingredients (grams, spoons, etc.) + short preparation steps.
    • Cultural adaptation (based on nationality or preference).
    • Clear macros (Proteins, Carbs, Fats, Calories) per meal.
  • ✅ Professional Validator → Reviews plan for nutritional logic, cultural fit, fasting compliance, clarity, and user applicability.
    • Provides decision: “Aprobado ✅” or “Ajustar ⚠️” with explanation.
• Conditional loop: nutritionist regenerates menus if validator requests adjustments.
• Outputs a final Markdown summary: coach insights + meal plans + validation + motivational closure.
• ✅ Interactive Gradio UI for InBody, goals, restrictions, and cultural context → structured multi-day nutrition plan.

---

📂 Repository Structure

• /examples → contains MVP agent notebooks (grow-pulse-daily-reader.ipynb, ai_daily_reader.ipynb, pura_vida_helper.ipynb).
• requirements.txt → Python dependencies.
• README.md → project overview and agent catalog.
• .gitignore → standard ignores.

(Planned: /prompts for centralized prompts, /api for FastAPI endpoints, /tests for agent tests.)

---

⚡ Tech Stack

• LangChain / LangGraph → agent orchestration.
• OpenAI (gpt-4o) → LLM backbone.
• Gradio → quick UI prototyping.

---

🛠️ Quickstart

# Create venv
python -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Run any notebook demo
jupyter notebook examples/grow-pulse-daily-reader.ipynb