Screen 10,000 Rows

Add the everyrow connector if you haven't already. Then upload the FDA product recalls CSV and ask Claude:

Screen this FDA product recalls dataset to find recalls of products that I might have used for my child born on 2021-08-01.

2,271 of 9,949 recalls are relevant (22.8%). Results take about 12 minutes.

Claude Code handles filtering a hundred rows natively by reading and evaluating each one. Scaling to 10,000 rows needs an approach where a fast pre-filter narrows candidates first, then LLM agents evaluate only the plausible matches individually.

Here, we get Claude Code to screen 9,949 FDA product recalls to find products relevant to a child born on 2021-08-01. Each row requires reasoning about whether a recalled product (food, medication, device) would plausibly be used for a child of that age.

Add everyrow to Claude Code if you haven't already:

claude mcp add futuresearch --scope project --transport http https://mcp.futuresearch.ai/mcp

With the FDA recalls CSV in your working directory, tell Claude:

Screen this FDA product recalls dataset to find recalls of products that I might
have used for my child born on 2021-08-01.

Claude calls everyrow's screen MCP tool. At this scale, the two-pass pipeline is critical: a fast first pass triages all 9,949 rows, then a careful second pass re-evaluates borderline cases with a stronger model:

Tool: everyrow_screen
├─ task: "Find recalls of products that I might have used for my child born on 2021-08-01."
├─ input_csv: "/Users/you/fda_product_recalls.csv"
└─ response_schema: null

→ Submitted: 9,949 rows for screening.
  Session: https://futuresearch.ai/sessions/310fc823-0adc-402c-bff1-7dc43fda2636
  Task ID: 310f...

Tool: everyrow_progress
├─ task_id: "310f..."
→ Running: 0/9949 complete (30s elapsed)

...

Tool: everyrow_progress
→ Completed: 9949/9949 (0 failed) in 707s.

Tool: everyrow_results
├─ task_id: "310f..."
├─ output_path: "/Users/you/child_relevant_recalls.csv"
→ Saved 2271 rows to /Users/you/child_relevant_recalls.csv

2,271 of 9,949 recalls are relevant. View the session.

Sample passing recalls (products a child could have been exposed to):

Sample non-passing recalls (correctly excluded):

The everyrow SDK's screen() function filters a dataframe by applying LLMs to every row. This demonstrates screening at scale: 10,000 FDA product recalls screened for personal relevance.

pip install everyrow
export EVERYROW_API_KEY=your_key_here  # Get one at futuresearch.ai/api-key

import asyncio
import pandas as pd
from everyrow import create_session
from everyrow.ops import screen

fda_recalls = pd.read_csv("fda_product_recalls.csv")
fda_recalls["center_classification_date"] = pd.to_datetime(
    fda_recalls["center_classification_date"], errors="coerce"
)
fda_recalls = fda_recalls[
    fda_recalls["center_classification_date"] > pd.Timestamp("2021-08-01")
]

async def main():
    async with create_session(name="FDA Recall Screening") as session:
        result = await screen(
            task="Find recalls of products that I might have used for my child born on 2021-08-01.",
            input=fda_recalls,
        )
        return result.data

results = asyncio.run(main())

At $0.001 per row, the cost scales linearly. The two-pass pipeline uses a fast model for initial triage and a stronger model for borderline cases, keeping accuracy high while controlling cost.