from learning to shipping
The Build Path 🛠️
you finished the chapters. this isn't another lesson — it's a practical checklist for turning what you learned into something real: picked, built, evaluated, deployed, and on your portfolio.
start here → pick a track, then work top to bottom, checking off each stage as you finish it.
pick a track in stage 1 to see this filled in ↑
pick a track in stage 1 to see this filled in ↑
eval template
test_cases = [
{"input": "...", "must_include": ["..."], "must_avoid": ["..."]},
# 4 more — cover the normal case, an edge case, and a case it should refuse
]
def run_eval(model_fn, test_cases):
passed = 0
for i, tc in enumerate(test_cases):
output = model_fn(tc["input"])
ok = all(s in output for s in tc.get("must_include", [])) and \
not any(s in output for s in tc.get("must_avoid", []))
print(f"test {i+1}: {'PASS' if ok else 'FAIL'}")
passed += ok
print(f"\n{passed}/{len(test_cases)} passed")streamlit
import streamlit as st
st.title("my project")
query = st.text_input("ask something")
if query:
with st.spinner("thinking..."):
answer = run_pipeline(query) # your rag/agent/model function
st.write(answer)next.js api route
// app/api/ask/route.js
export async function POST(req) {
const { query } = await req.json();
const answer = await runPipeline(query); // your rag/agent/model function
return Response.json({ answer });
}Streamlit Community Cloud
freethe fastest path for a streamlit app — connect a github repo, click deploy.
watch out: public repo required on the free tier; app sleeps after inactivity and takes a few seconds to wake up.
Vercel
free tierthe natural home for a next.js app — push to git, it deploys itself, instant preview URLs.
watch out: serverless functions have execution time limits — fine for API calls, watch out for slow local model inference.
Hugging Face Spaces
free (CPU)built for exactly this — supports streamlit, gradio, or a raw docker container. natural fit if your project already lives around HF models.
watch out: free tier is CPU-only; anything GPU-hungry needs a paid space.
a small VPS
~$4-6/mofull control — long-running processes, local model inference, no platform limits.
watch out: you're the ops team now: updates, uptime, and a firewall are on you.
0/8 stages checked off — keep going.