Ship 12
The Agentic AI Builder Program
You don't watch tutorials — you ship products. In 12 weekends you build, test and publicly deploy 12 production-grade AI products, each directed by AI agents and anchored to a real system-design case study from companies operating at massive scale.
A portfolio the world can actually open
Most courses end with a certificate and a notebook nobody runs. Ship 12 ends with twelve working products on the public internet — built the way modern engineering teams work.
12 deployed links
Real apps, real URLs. Anyone can open them — recruiters, friends, hiring managers.
GitHub portfolio
Prioritized issues, agent-authored PRs, reviews and green CI — the artifacts hiring managers want.
System-design fluency
Argue SQL vs NoSQL, pick queues and vector DBs, and defend tradeoffs across 12 architectures.
Agent-native workflow
Take any idea and ship it by directing AI agents — not hand-writing every line.
Certificate
Issued once all 12 products are deployed and reviewed by mentors.
Career support
Resume building around your 12 projects, mock interviews and referral guidance.
Become the architect. The agent becomes the implementer.
Why agentic, why now.
You own the decisions
You brainstorm, validate, design, write requirements and review every change. The agent writes the code.
No syntax grind
Your energy goes to data models, scaling and correctness — what actually matters.
You learn real systems
Every build is paired with a published case study, so you understand why systems look the way they do.
One build methodology, twelve products
Every week runs the same end-to-end loop. Repeat it across 12 domains and the process becomes muscle memory — so you can ship anything after the program.
Brainstorm the idea
Pick the product. Build intuition: who it's for, what problem, why it matters.
Validate it
Pressure-test it. Is it real? Can it ship in one weekend? Smallest useful version?
System-design brainstorm
Whiteboard the architecture before any code. Debate SQL/NoSQL, queues, vector DB, tools.
Gather requirements
Capture functional & non-functional reqs — latency, scale, cost — AI-assisted.
Agent plans the work
The agent converts requirements into prioritized GitHub issues (P0/P1/P2).
Agent builds, you direct
Agent picks the top issue, implements, opens a PR. You review, steer, approve. No hand-coding.
Manual test & QA
You test every feature against the plan. Gaps become new issues for the agent.
Deploy publicly
Ship to a public URL. Capture the live link for your portfolio.
Demo + retro
Show it live. Reflect: which tradeoff paid off, what changes at 100× scale.
Make & defend real tradeoffs — a dozen times
Before any code is written, the cohort runs a system-design brainstorm with a mentor. By Week 12 you'll have made each of these decisions in a dozen contexts — exactly what interviews and senior roles test.
Data store
SQL vs NoSQL? Relational integrity vs flexible schema?
Vector store
pgvector vs Pinecone vs FAISS? When is a dedicated DB worth it?
Queue / async
Celery, Redis Streams, SQS or Kafka — which, and why?
Caching
What to cache, where (Redis?), and how to invalidate it.
Batch vs stream
Precompute offline or score in real time? Latency budget?
Model serving
Hosted API vs self-hosted? Cost vs control vs latency.
Observability
Logs, metrics, error tracking — is prod healthy?
Cost & scale
What breaks at 10× and 100×? Cheapest correct design?
Guided by engineers from big tech
Mentors currently working at top companies lead the system-design brainstorms, run the case-study spotlights, and review your issues, agent PRs and deployments — at a small ≤ 1:15 mentor-to-student ratio.
12 weeks · 12 shippable products
One publicly deployed product per week, anchored to a real engineering case study. Built Python-first: FastAPI, Claude Agents SDK / LangGraph, Postgres + pgvector, Redis, Docker, deployed on Render / Fly.io / Railway.
RAG Support / FAQ Agent
⌁ arXiv QA-with-RAG · GitHub Enterprise LLM
A chatbot answering questions over a document set with citations and graceful “I don't know.”
Visual “Search by Image” Engine
⌁ Etsy — Search by Image
Upload an image, get visually similar items back via multimodal embeddings.
ETA / Delivery-Time Prediction Service
⌁ Swiggy — ETA Modeling
An API predicting ETA from distance, time-of-day and load, with prediction logging.
Personalized Recommender
⌁ Netflix Recsys · Dailymotion Vector Recsys
“Users who liked X also like…” recommendations that refresh on new interactions.
GenAI Content-Moderation Pipeline
⌁ Whatnot — GenAI Trust & Safety
Flags unsafe content with explanations and routes edge cases to human review.
Entity Resolution / Dedup Service
⌁ Walmart — Entity Resolution Framework
Detects that “Acme Inc.”, “ACME Incorporated” and “acme” are the same entity.
Fraud / Anomaly Detection Service
⌁ Grab Graph Anomaly · Uber Risk Entity
Flags suspicious transactions/entities and shows why they were flagged.
Price-Alert / Event-Driven Monitor
⌁ Expedia — Price Alerts
Watch a price/metric and notify users when a threshold is crossed.
Customer LTV / Churn Predictor + Dashboard
⌁ Expedia — Customer LTV Prediction
Predict lifetime value / churn risk with a visual segment dashboard.
Embedding Job / Lookalike Matcher
⌁ Grab Lookalikes · LinkedIn Job Matching
Match candidates ↔ jobs (or build a “lookalike” audience) via embeddings.
Multi-Agent GenAI Product
⌁ DoorDash GenAI · Salesforce Einstein Search
An assistant orchestrating multiple specialized agents and tools to complete a task.
Capstone + Scale Topic — Demo Day
⌁ Stitch Fix Distributed Training · Meta Code Llama
A student-chosen product combining earlier skills, ending in a live demo-day presentation.
Anatomy of a 3.5-hour session
Sessions are interactive build time, not lectures. Here's the minute-by-minute shape of a session (≈210 minutes).
Standup + case-study spotlight
Review last build; a mentor spotlights the week's real-world case study and its scale numbers.
System-design brainstorm
Whiteboard the architecture; debate SQL/NoSQL, queue, vector DB, caching and tools — capture every tradeoff.
Idea validation + requirements
Validate scope for one weekend; gather functional + non-functional requirements, AI-assisted.
Agent plans the work
The agent converts requirements into prioritized GitHub issues (P0/P1/P2).
Agentic build
The agent picks issues by priority and implements; you monitor, steer and review diffs/PRs — no hand-coding.
Manual test & QA
Test each feature against the plan; gaps become new issues for the agent.
Deploy + demo + retro
Deploy to a public URL, capture the live link, quick retrospective.
🟣 Saturday · Session 1
Standup → case study → system-design brainstorm → idea validation → requirements → agent creates prioritized issues → start the build. By Saturday evening the architecture is decided and the first PRs are merged.
🟠 Sunday · Session 2
Continue the agentic build → manual testing against the plan → deploy publicly → demo → retro. The product ships by Sunday evening and the live link goes into your portfolio.
30 real-world systems that anchor the program
Each weekly spotlight draws from this library, so you learn how the same problems are solved at companies operating at insane scale — then build a hobby-scale version yourself.
Recommendations & Personalization
GenAI · RAG · LLM Apps
Forecasting · ETA · Pricing
Data Quality · Entity Resolution · Inventory
Anomaly · Risk · Fraud
Search · Media · Infra
Built for builders, no degree required
Light prerequisites, real outcomes. If you can write basic Python and you're hungry to ship, you're ready.
Who it's for
- Students & early-career engineers
- Career switchers into AI/ML
- System-design interview prep
- Self-taught devs tired of tutorial hell
Prerequisites
- Basic Python
- Terminal & Git basics
- Laptop (8GB+ RAM), stable internet
- GitHub + Claude account (free works)
NOT required
- A CS degree
- Prior ML / deep-learning experience
- Frontend / DevOps expertise
- Hand-writing production code
Format & logistics
- Weekend batch — Sat + Sun, ~3.5h each
- ~7h/week live · ~84h total
- Live online / hybrid · small cohort
- ≤ 1:15 mentor ratio · recordings included
Your stack
- Python · FastAPI · Docker
- Claude Agents SDK / LangGraph
- Postgres + pgvector · Redis
- GitHub Issues/Actions · Render/Fly.io
Ship 12. Walk away with 12 live links.
12 weeks · 12 deployed products · a ByteStackOne weekend program. Not a transcript — twelve things the world can actually open and use.
One-time enrollment fee, shared when your application is accepted. Tools run on free / hobby tiers; you bring your own Claude usage (free tier works to start). No markup — you pay Anthropic directly.