micro1 Agentic Workflows Hackathon submission
# Field Data Quality Agent
### micro1 Agentic Workflows Hackathon — Aug 2026
## 1. Who has this problem?
M&E (Monitoring & Evaluation) officers and field supervisors at NGOs running
household beneficiary surveys — for cash transfer, food aid, or general needs
assessment programs. This is common in humanitarian operations across Northeast
Nigeria and similar contexts, where dozens of enumerators submit household survey
data from the field, often under time pressure and with inconsistent connectivity.
## 2. What bottleneck makes it worth solving?
Before beneficiary data can be used for targeting or reporting, someone has to
review it for quality. Today this is either skipped, done with a simple script
that checks whether individual numbers fall in a valid range, or done by hand.
Both approaches miss the errors that actually matter most:
- **Numbers that are individually valid but don't add up together** (household
size doesn't match the number of adults + children reported).
- **Numbers that contradict the enumerator's own free-text notes** (income far
too high for a household described as "subsistence farming, no other income").
- **Patterns across multiple records** that suggest fabricated or duplicated
entries — e.g. one enumerator submitting five near-identical household surveys
within nine minutes, which is not a data-entry typo but a sign the visits may
not have genuinely happened.
A simple range-check script catches obvious typos (age = 200) but is structurally
blind to all three of the above, because each of them requires *relating* several
pieces of information to each other rather than checking one number in isolation.
Missing these errors means aid targeting or program reporting is built on data
nobody actually verified was real.
## 3. Why this is valuable in practice
Catching these errors before data reaches a report or a targeting decision protects
both program integrity (donor trust, audit risk) and the beneficiaries themselves —
a mis-recorded hous …