How to use Devin to batch CRM duplicate candidates
Devin batched CRM duplicate candidates, preserved pair cursors and archived-chain behavior, and merged a patch with a scoped production query measurement.
github.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Optimize this duplicate scan’s database round trips without changing pair decisions. Batch candidate retrieval, keep blocking keys a superset of current match rules, and preserve mid-page resume cursors. Add equivalence and archived-merge-chain regressions, report scoped query measurements, and prepare a reviewed pull request before release.
Reach people exploring AI agents
See placements →What happened
Merged PR 74 batches duplicate candidates and preserves resume and merge semantics; its body reports a scoped 224 ms candidate query rather than a whole-scan benchmark.
Reported by devin-ai-integration[bot], requested by @Reichenbachian — GitHub pull requestHow the agent did it
- The requesting human delegated remaining duplicate-scan round-trip costs to Devin.
- Devin batched contact peers and company blocking keys while retaining the existing final matching rules.
- The patch added cursor-resume and archived-chain regression coverage and was merged into the release branch.
Where a person stepped in
The PR names @Reichenbachian as requester and was merged; the public evidence does not expose the private session prompt or detailed human review conversation.
A StructifyAI CRM duplicate scan still processed too few contacts per scheduled tick after an earlier query optimization. The PR describes the remaining cost as database round trips and repeated checkpoint writes, with company comparisons also traversing a quadratic candidate space.
Devin batched contact candidates for a page of 50 sources, using shared email, LinkedIn and company evidence, then loaded peer records together. Company blocking keys formed a superset of existing match rules; the original comparison functions still made final decisions. This preserved matching semantics while reducing unnecessary retrieval.
The change also preserved a precise pair cursor when a tick ended mid-page and tracked records archived by merges earlier in that tick. Those details matter because faster batching must not repeat merges or skip unfinished pairs. Regression coverage checked page-boundary equivalence, resumption and an archived merge chain.
The PR reports a production query measurement of 224 milliseconds for a real 50-contact candidate page, compared with approximately ten seconds of earlier round trips. That is a scoped measurement, not a measured speedup for the entire scheduled scan. The bot-authored PR links its Devin session, names the requesting human and is verified merged. No private session transcript or broader production benchmark was inspected.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: GitHub pull request · devin-ai-integration[bot], requested by @Reichenbachian
Frequently asked questions
Did batching replace the final duplicate rules?
No. Blocking keys generate candidates, while the existing matching checks still make the decision.
How did the patch handle interrupted pages?
It preserved an exact pair cursor so the next tick resumes unfinished work.
What performance result is actually reported?
A real 50-contact candidate query took 224 ms; an entire post-change scan duration is not supplied.
More details
About this account
The requesting human delegated remaining duplicate-scan round-trip costs to Devin. Devin batched contact peers and company blocking keys while retaining the existing final matching rules.
Opened public PR through GitHub connector, including diff, author, timestamps and merged state. Agent session link and requester establish attribution. Performance figures come from the PR body; private execution logs were not inspected. Suggested prompt written by Agents In Use; no original task prompt was found in the verified text.
Credit: devin-ai-integration[bot], requested by @Reichenbachian — GitHub pull request
Source published 14 Oct 2026
Source checked 14 Oct 2026
Tools and topics
Devin · PostgreSQL · Prisma · GitHub
database round trips · duplicate detection · resume cursor
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.