AI for professionals in Lima

1. Name one task. Pick a weekly task that costs you an hour: a report, a reconciliation, a set of replies.
2. Do it once with AI, by hand. In class you use an assistant on that task and note every place it is wrong or vague.
3. Write the checks. Before automating anything, decide how you will know the output is right: a total that must match, a field that must be present.
4. Automate the safe part. Only then turn it into a script, a spreadsheet flow or an agent, with the checks built in.
5. Keep a person in the loop. Decide which outputs a human approves before they leave the building.

Professionals who want the national picture, including time zones and checking AI output in structured formats, can read AI and coding for professionals in Peru. Teams should see the Spanish curso de IA para empresas, and those focused on agents the curso de agentes de IA para profesionales.

import csv
from collections import Counter
from datetime import datetime

per_year = Counter()
with open("AI_0626.csv", encoding="latin-1") as raw:
    for row in csv.DictReader(raw, delimiter=";"):
        try:
            made = datetime.strptime(row["FECHA_CREACION"], "%Y%m%d%H%M%S")
        except ValueError:
            continue                      # the draft's quiet "fix"
        per_year[made.year] += 1

print("Requests counted:", sum(per_year.values()))
print(dict(sorted(per_year.items())))

The result says the municipality received its first requests in 2025. But the file has 25,415 rows and its own year column runs from 2021. Four rows in five vanished without a message, because the except clause swallowed every error. Version 2 refuses to skip anything: it labels each timestamp by how precise it is and counts the labels.

import csv
from collections import Counter
from datetime import datetime

def precision(stamp):
    try:
        datetime.strptime(stamp, "%Y%m%d%H%M%S")
        return "full"
    except ValueError:
        if stamp[6:] == "00000000":
            return "month only"
        return "unreadable"

kinds = Counter()
by_year = Counter()
with open("AI_0626.csv", encoding="latin-1") as raw:
    for row in csv.DictReader(raw, delimiter=";"):
        kind = precision(row["FECHA_CREACION"])
        kinds[kind] += 1
        by_year[(row["FECHA_CREACION"][:4], kind)] += 1

print("Rows by precision:", dict(kinds))
for year in ("2021", "2023", "2025", "2026"):
    print(year, "full", by_year[(year, "full")], "month only", by_year[(year, "month only")])

Data: Municipalidad Metropolitana de Lima, Solicitudes de acceso a la información pública, 2021 to 2026, national open-data portal, ODC-By licence; the counts are ours. Until January 2025 the file records only the year and month, written like 20210700000000 with day and time set to zero; full timestamps start on 1 February 2025.

Nothing in the data is wrong as such. The register simply recorded only the month for its older entries, and the "00" day is how it says so. The mistake was the fix. A professional's habit, which we drill in class, is to reconcile: before trusting a count, compare it with a total you already know, here 25,415 rows, and make every rejected row show up in a labelled bucket. With the full stamps you can then answer finer questions honestly, such as how many requests arrived at weekends since February 2025: 105 of 5,135.

1. Prompt better. Ask an AI assistant to fix Version 1 without discarding rows, and to report how many rows each branch handles. Compare its answer with Version 2.
2. Reconcile. Add a final check that the precision counts add up to the number of rows read, and stop the script if they do not.
3. Spreadsheet version. Import the file into Excel or Sheets and find the same problem with a filter on the day digits.
4. Choose the right grain. Count requests per month for all six years, which every row supports, instead of per day.
5. Explain it. Write three sentences, in English, for a manager who saw the draft's table and wants to know what changed.

Adults who started later in their careers often prefer a slower pace; coding for adults over 50 in Peru describes how we adapt. Analysts who want to go deeper into queries can follow the SQL and data analysis course for Peru.

Lima stays on UTC minus 5 hours without summer time (tz database, America/Lima), India on UTC plus 5 hours 30 minutes, a constant gap of 10 hours 30 minutes.

You book in Lima time and the mentor sees India time. Because neither clock changes, the hour you choose in January is still the hour in July, which makes it easy to block in a work calendar.