Python politely told me to stop writing JavaScript in it

This is post 3 of my TypeScript-to-Python journey. Old habits die hard however, Python keep telling me to stop writing JavaScript in it.

Quick background: I am a full-stack developer expanding into AI engineering, learning Python by building GenAI apps directly. Today was about the everyday plumbing: filtering data and assembling prompts, which is 70% of what LLM application code does.

My instinct after ten years of TypeScript:

jobs.filter((j) => j.remote).map((j) => j.title);

Python has map() and filter() too, but the community considers them unidiomatic. The Pythonic way is a comprehension: one expression that does both.

[j["title"] for j in jobs if j["remote"]]

It reads strangely for the first hour. Very counter-intuitive. Until i realized what is it about and how to wrap my head around it.

[ j["title"]   for j in jobs   if u["remote"] ]
  ^ .map(...)  ^ the iteration   ^ .filter(...)

Notice the reading order of a comprehension: EXPRESSION, then for-loop, then filter. "Give me j["title"], for each j in jobs, if job is remote." The filter comes last even though it runs first.

Two things very interesting thing that I learnt:

1. join() is reversed. It is not list.join("\n"), it is "\n".join(list). The separator string owns the method, not the list.

TypeScript

list.join("\n");

Python

"\n".join(list)

2. Missing dict keys throw KeyError instead of returning undefined. So config.get("model", fallback) is the Python way of config.model ?? fallback.

TypeScript

config.model ?? fallback;

Python

# this will trow `KeyError` if key does not exists
config["model"]
# Even without fallback get will return NONE
config.get("model", fallback)

dict["key"] raises KeyError if the key is missing. dict.get("key", fallback) is the safe version, with the fallback as the second argument instead of a ?? operator.

f-strings are python way of template literals, it beats template literals for formatting. f"{salary:,}" prints 195,000 with the thousands separator baked in, no library needed.

salary = 195000
f"{salary:,}"  # "195,000"

Template literals cannot do that on their own.

Win today

A script that loads AI-engineer job postings from JSON, filters the remote Python ones with a comprehension, and assembles a formatted prompt using f-strings. Small thing, but this is the exact shape of prompt assembly inside every RAG pipeline, so I will be reusing this muscle weekly.

I created a jobs.json

[
  {
    "title": "GenAI Application Engineer",
    "company": "Acme AI",
    "salary_max": 185000,
    "remote": true,
    "skills": ["python", "fastapi", "rag", "evals"]
  },
  {
    "title": "ML Platform Engineer",
    "company": "DataCo",
    "salary_max": 210000,
    "remote": false,
    "skills": ["python", "kubernetes", "mlops"]
  },
  {
    "title": "AI Full-Stack Developer",
    "company": "StartupX",
    "salary_max": 165000,
    "remote": true,
    "skills": ["typescript", "python", "react", "llm-apis"]
  },
  {
    "title": "Frontend Engineer",
    "company": "WebShop",
    "salary_max": 140000,
    "remote": true,
    "skills": ["typescript", "react", "css"]
  },
  {
    "title": "LLM Engineer",
    "company": "PromptLabs",
    "salary_max": 195000,
    "remote": true,
    "skills": ["python", "prompt-engineering", "evals", "agents"]
  },
  {
    "title": "Data Engineer",
    "company": "PipeCorp",
    "salary_max": 170000,
    "remote": false,
    "skills": ["python", "sql", "airflow"]
  }
]

Then wrote a script to create the prompt for the AI


import json
from pathlib import Path

postings = json.loads(Path("jobs.json").read_text())

matches = [
    job for job in postings
    if job["remote"] and "python" in job["skills"]
]

# sorted(key=...) is your arr.sort(compareFn); [:3] is .slice(0, 3)
top = sorted(matches, key=lambda job: job["salary_max"], reverse=True)[:3]

# .map(...) to formatted lines; enumerate replaces the (x, i) callback
lines = [
    f"{i}. {job['title']} at {job['company']}"
    f" (up to ${job['salary_max']:,})"
    f" - skills: {', '.join(job['skills'])}"
    for i, job in enumerate(top, start=1)
]
jobs_block = "\n".join(lines)

prompt = f"""You are a career coach for a full-stack TS developer
moving into AI engineering.

Here are their top {len(top)} matching job postings:

{jobs_block}

Based only on these postings, which single skill should they
learn next? Answer in two sentences."""

print(prompt)

Output

You are a career coach for a full-stack TS developer moving into AI Engineering.

Here are their top 3 matching job postings:

1. LLM Engineer at PromptLabs (up to $195,000) - skills python, prompt-engineering, evals, agents
2. GenAI Application Engineer at Acme AI (up to $185,000) - skills python, fastapi, rag, evals
3. AI Full-Stack Developer at StartupX (up to $165,000) - skills typescript, python, react, llm-apis

based only on these postings, which single skill should they learn next?

Answer in two sentences

My takeaway

Syntax is easy, unlearning hard. The second part is where the actual work is.

🖖🏼 References

FastAPI is the Express + Zod + Swagger stack collapsed into the type system

The last post in the series. I rebuilt my Express reflexes in FastAPI, found the type hints doing the validation, the docs, and the serializing, then made it stream.

Python FastAPI Gen AI Python for TypeScript devs

TS➞PYPython for TypeScript devs

#04

You start it

async/await, but the event loop is not ambient anymore

A full-stack dev learning AI engineering in public

async def learn():

20 Jul 2026•Vikas Yadav

Python handed me the keys to the event loop and said you start it

async/await transfers almost 1:1 from JS, but Python's event loop is not ambient. Two gotchas from wiring up token-by-token streaming.

Python Async Gen AI Python for TypeScript devs

TS➞PYPython for TypeScript devs

#02

Zod taught me Pydantic

Define the schema, parse at the boundary. Same mental model, new language.

A full-stack dev learning AI engineering in public

async def learn():

13 Jul 2026•Vikas Yadav

Pydantic: Zod for Python

If you have used Zod, you already understand 70 to 80% of Pydantic. The mental model transfers almost one-to-one.

Python Pydantic Gen AI Python for TypeScript devs

[ { "title": "GenAI Application Engineer", "company": "Acme AI", "salary_max": 185000, "remote": true, "skills": ["python", "fastapi", "rag", "evals"] }, { "title": "ML Platform Engineer", "company": "DataCo", "salary_max": 210000, "remote": false, "skills": ["python", "kubernetes", "mlops"] }, { "title": "AI Full-Stack Developer", "company": "StartupX", "salary_max": 165000, "remote": true, "skills": ["typescript", "python", "react", "llm-apis"] }, { "title": "Frontend Engineer", "company": "WebShop", "salary_max": 140000, "remote": true, "skills": ["typescript", "react", "css"] }, { "title": "LLM Engineer", "company": "PromptLabs", "salary_max": 195000, "remote": true, "skills": ["python", "prompt-engineering", "evals", "agents"] }, { "title": "Data Engineer", "company": "PipeCorp", "salary_max": 170000, "remote": false, "skills": ["python", "sql", "airflow"] } ]

import json from pathlib import Path postings = json.loads(Path("jobs.json").read_text()) matches = [ job for job in postings if job["remote"] and "python" in job["skills"] ] # sorted(key=...) is your arr.sort(compareFn); [:3] is .slice(0, 3) top = sorted(matches, key=lambda job: job["salary_max"], reverse=True)[:3] # .map(...) to formatted lines; enumerate replaces the (x, i) callback lines = [ f"{i}. {job['title']} at {job['company']}" f" (up to ${job['salary_max']:,})" f" - skills: {', '.join(job['skills'])}" for i, job in enumerate(top, start=1) ] jobs_block = "\n".join(lines) prompt = f"""You are a career coach for a full-stack TS developer moving into AI engineering. Here are their top {len(top)} matching job postings: {jobs_block} Based only on these postings, which single skill should they learn next? Answer in two sentences.""" print(prompt)

You are a career coach for a full-stack TS developer moving into AI Engineering. Here are their top 3 matching job postings: 1. LLM Engineer at PromptLabs (up to $195,000) - skills python, prompt-engineering, evals, agents 2. GenAI Application Engineer at Acme AI (up to $185,000) - skills python, fastapi, rag, evals 3. AI Full-Stack Developer at StartupX (up to $165,000) - skills typescript, python, react, llm-apis based only on these postings, which single skill should they learn next? Answer in two sentences

FastAPI is the Express + Zod + Swagger stack collapsed into the type system

The last post in the series. I rebuilt my Express reflexes in FastAPI, found the type hints doing the validation, the docs, and the serializing, then made it stream.

Python FastAPI Gen AI Python for TypeScript devs

TS➞PYPython for TypeScript devs

#04

You start it

async/await, but the event loop is not ambient anymore

A full-stack dev learning AI engineering in public

async def learn():

20 Jul 2026•Vikas Yadav

Python handed me the keys to the event loop and said you start it

async/await transfers almost 1:1 from JS, but Python's event loop is not ambient. Two gotchas from wiring up token-by-token streaming.

Python Async Gen AI Python for TypeScript devs

TS➞PYPython for TypeScript devs

#02

Zod taught me Pydantic

Define the schema, parse at the boundary. Same mental model, new language.

A full-stack dev learning AI engineering in public

async def learn():

13 Jul 2026•Vikas Yadav

Pydantic: Zod for Python

If you have used Zod, you already understand 70 to 80% of Pydantic. The mental model transfers almost one-to-one.

Python Pydantic Gen AI Python for TypeScript devs

Python politely told me to stop writing JavaScript in it

Win today

Output

My takeaway

🖖🏼 References

Read next

FastAPI is the Express + Zod + Swagger stack collapsed into the type system

Python handed me the keys to the event loop and said you start it

Pydantic: Zod for Python

Python politely told me to stop writing JavaScript in it

Win today

Output

My takeaway

🖖🏼 References

Read next

FastAPI is the Express + Zod + Swagger stack collapsed into the type system

Python handed me the keys to the event loop and said you start it

Pydantic: Zod for Python