One Line of User Feedback, Fixed Within a Day. MFPRO Search Upgraded + Complete AMFI NAV History, 37M Rows, 2006 Onward.
Published: July 18, 2026
Someone typed one line into the Feedback box on MFPRO yesterday evening, "ABSL Value fund and ICICI Pru Value Fund are missing". Whoever you are, thank you, the fix went live within a day. With enhancements, including expanded search options.
The funds were in the database all along, the bug was in the search, and while at it the search got a proper upgrade, it understands ABSL and PPFAS and old fund names, and it ranks results by relevance rather than alphabet.
The second one - some days back users were hitting 404s on schemes that should have existed. I read every error in my logs, every day, and those 404s turned into a full rebuild, every NAV AMFI has ever published, 2006 onward. Now includes closed, defunct, interval & close ended funds also.
Now from a single platform - run advanced analytics on the interface, access data via APIs, download the full database, 37M rows, in one shot, and even get Python codes to compute metrics offline.
For the methodology and the step by step validations, it is all in the Docs section on MFPRO.
More on the search solution
The problem
We tend to type short names, the database stores long official ones. You type ABSL Value, the record says Aditya Birla Sun Life Value Fund, and a literal search finds nothing. Same story when a fund house renames itself, you remember Reliance Small Cap, the fund is now Nippon India Small Cap.
The fix has two parts
One, the search now keeps a list of these short forms and old names, and quietly expands them. Type ABSL, it searches Aditya Birla Sun Life. Type IDFC Flexi, it finds Bandhan Flexi Cap, the fund that renamed in 2023. Type PPFAS, it finds Parag Parikh. Seventeen mappings so far, each one added because a real search failed.
Two, results are scored instead of sorted alphabetically. A tighter match to what you typed scores higher, actively tracked funds score higher, so the fund you meant sits at the top instead of twenty index variants that happen to share a word.
Will continue to modify the rules based on feedback and my own experience using the tool.
The validation ran at every stage rather than once at the end
At the pull: the full AMFI archive was downloaded and profiled offline first, nothing touched the main database at this stage.
At the parse: scheme category in the old archives lives in section header lines rather than on each record, so the parser carries that context through. Legacy category names were mapped to the current taxonomy through an explicit mapping I reviewed and approved, no silent guessing.
At the load: a pilot batch first, then date-ranged batches, and after every batch a cross-tab by year and scheme type, before plus added equals after, checked cell by cell before the next batch ran. The grand total closes to the row.
Existing rows always win, and the nine conflicting overlaps found in 16.4 million rows are logged and documented.
After the load: a separate multi-agent audit re-checked the whole chain, seven agents finding issues, seven adversarial agents verifying them. Zero duplicate scheme and date pairs across 37 million rows, zero orphans. Real items it caught were fixed the same day.
And the source's own oddities, like zero-NAV rows in some 2006 era files, are published as filed in the docs rather than silently cleaned.
