Skip to content

Instantly share code, notes, and snippets.

@alexhawkins

alexhawkins/level-up-with-ai.md

Last active August 14, 2025 09:59
Show Gist options
  • Save alexhawkins/0be601c474bf1b8b977b7dbedfbf52b1 to your computer and use it in GitHub Desktop.
Save alexhawkins/0be601c474bf1b8b977b7dbedfbf52b1 to your computer and use it in GitHub Desktop.
Download ZIP
📚 Ultimate AI Marketing, Analytics & Agentic Workflows — Beginner to Pro (10/10)
Raw
level-up-with-ai.md

📚 Ultimate AI Marketing, Analytics & Agentic Workflows — Beginner to Pro (10/10)

By Alex — for Pear

🎯 End Goals

  • Build AI apps with OpenAI, Claude, Gemini
  • Generate images/videos for ads and socials at scale
  • Use AI for analytics and charting
  • Ship multi-agent automations for marketing & ops
  • Integrate with Zoom, Gmail, Notion/DBs, CRMs
  • Deploy safe, cost-efficient, production-ready systems

🗓 Week 0 — Foundations

Learn

  • LLM basics (tokens, embeddings, context, hallucination)
  • APIs & JSON, environment variables, Git/GitHub
  • Tooling: VS Code, Node or Python

Mini-Wins

  • Call a public API (Open-Meteo), print JSON, commit to GitHub

🗓 Week 1 — First LLM Calls

Docs

Mini-Wins

  1. Generate a launch tweet thread (OpenAI)
  2. Summarize a blog post (Claude)
  3. Produce 3 brand tones for a product card (Gemini)

🗓 Week 2 — Image Generation for Marketing

Docs

Mini-Wins

  • Generate 5 ad variants + 3 copy angles
  • Create multi-size social banners (1:1, 16:9, 9:16)

🗓 Week 3 — RAG & Analytics

Docs

Mini-Wins

  • Upload CSV, ask: “Which ads have the best CAC this month?”
  • Auto-chart “revenue by month” → PNG/HTML chart

🗓 Week 4 — Multi-Agent Marketing Bots

Docs

Roles

  • Planner → Copywriter → Designer (images) → Scheduler (posts)

🗓 Week 5 — Claude Code & Automation

Docs

Mini-Wins

  • Map repo, find dead code
  • Auto-draft tests for 1 module, review before apply

🗓 Week 6 — Full-Stack AI Integration

Docs

Mini-Wins

  • MCP server exposing analytics table
  • Route to cheapest/fastest model with LiteLLM

🗓 Week 7 — Safety, Scaling & Deployment

Docs

Mini-Wins

  • Guardrail for prompt injection
  • Cost/latency tracking in Langfuse

🚀 Capstone Projects (Traffic & Revenue Focus)

Each project includes stack, milestones, acceptance criteria, and “level-up” ideas.

Project A — Ship a Mobile App from Scratch with Claude Code

Outcome: A cross-platform mobile app that fetches products, generates ad copy/images, and schedules social posts.

Stack

  • Frontend: React Native (Expo)
  • Backend: Next.js (API routes) or FastAPI
  • AI: Claude 3.7/3.5 for planning/copy, OpenAI gpt-image-1 for images
  • Router/Obs: LiteLLM + Langfuse

Claude Code Use

Milestones

  1. App shell + auth
  2. “Ideas → Copy → Images” flow
  3. Post scheduler UI (stores posts + timestamps)
  4. Deploy (TestFlight/Play Internal)

Acceptance

  • Create a campaign with 3 copy variants
  • Render 3 images per copy variant
  • Queue posts for 2 channels

Level-Up

Project B — Image & Video Pipeline: Kling + GPT-Image-1 + Veo 3 (+ Sora*)

Outcome: A prompt-driven pipeline that generates stills and short videos for ads.

Docs

Milestones

  1. Images: Generate 5 ad stills per product with gpt-image-1
  2. Video from text: 8s 720p with Veo 3 via Gemini API
  3. Video from image: Image-to-video with Veo 3 (or Kling if available)
  4. Packaging: Export bundles per channel (TikTok, Reels, YT Shorts)
  5. Optional: If you have Sora access, add a Sora job queue

Acceptance

  • Each product → 3 image styles + 1 video cut
  • All assets named with campaign, size, and platform
  • A JSON manifest listing files, captions, CTAs

Notes

Project C — Zoom Auto-Responder Bot (Team Chat + Meeting Chat)

Outcome: A Zoom bot that watches chat, answers FAQs, and hands off to a human when confidence is low.

Docs

Stack

  • Bot service (Next.js API or Cloudflare Worker), LiteLLM router, RAG over FAQ docs

Milestones

  1. App registration + OAuth
  2. Receive chat events → classify intent → answer with sources
  3. Low-confidence → tag human + log to inbox
  4. Add slash commands: /faq, /status, /hand-off

Acceptance

  • Responds in <2s for cached/FAQ answers
  • Includes source snippets for factual replies
  • Logs all turns to Langfuse

Project D — Gmail Reply Agent (Sales/Support)

Outcome: Triages inbound Gmail, drafts smart responses, and queues send for human approval (or auto-send by rule).

Docs

Flow

  • Gmail webhook → classify (lead, support, billing, spam) → draft with GPT → include 2 variants + confidence → auto-send if rule matched, else queue in “Drafts”

Milestones

  1. OAuth + messages.list + messages.get
  2. Categorizer + draft generator (tone control)
  3. messages.send for approved drafts
  4. Safety: no links for first-time senders, PII scrub

Acceptance

  • Triage accuracy ≥ 90% on labeled sample
  • First draft in < 4s, includes short + long variant
  • Zero-shot PII redaction on outgoing mail

Project E — Marketing Campaign Generator (E2E)

Outcome: One-click campaign planning → copy → images → schedule.

Tools

  • Copy: OpenAI GPT
  • Images: gpt-image-1
  • Scheduler: Buffer or native social APIs
  • Router & obs: LiteLLM + Langfuse

Acceptance

  • At least 2 platforms (IG + TikTok)
  • A/B copy + A/B images per platform
  • Calendar view + manual override

Project F — Sales & Marketing Insights Bot

Outcome: Upload sales/ads CSV → ask questions → get charts.

Milestones

  1. CSV ingestion → DuckDB table
  2. NL → SQL → Plotly chart (line/bar/scatter)
  3. Weekly digest email with highlights

Acceptance

  • Answers “top 5 ads by ROAS last 30 days”
  • Exports PNG chart + CSV slice
  • Tracks cost/time per query

⚡ Performance & Cost Optimization

  • Use small/fast models (e.g., GPT-4o-mini/Haiku) for drafts; reserve larger models for final QA
  • Cache prompts/results; batch image/video jobs
  • Preprocess (regex, SQL, vector search) before asking the LLM
  • Add Prompt Caching and tracing; set rate limits per user/session

📦 Deployment Recipes

  • Vercel: API routes + Vercel AI SDK UI
  • Render / Fly: long-running agents or queues
  • Supabase: auth + pgvector
  • Cloudflare Workers/Queues: low-latency webhooks and schedulers

đź“‚ Practice Datasets

  • marketing_analytics.csv (campaign, impressions, CTR, CPC, CPA, ROAS)
  • sales.csv (sku, price, units, channel, region, date)
  • seo_keywords.csv (keyword, vol, difficulty, CPC, intent)

🏎 10-Hour Speedrun (Recap)

  1. 2h Gemini quickstart + CLI chat oai_citation:26‡Google AI for Developers
  2. 2h LangChain RAG tutorial
  3. 2h Planner→Researcher→Writer with LangGraph
  4. 2h Claude Code repo mapping/tests oai_citation:27‡Anthropic
  5. 2h Promptfoo red-team + minimal deploy

đź”— Handy Links

OpenAI

Anthropic

Google / Veo

Sora (access-gated)

Zoom

Gmail

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment