Kishan ,
While trying to get SearchMyBlogs to draft a blog in respect of “ achievements – announcements – schemes “ mentioned by PM Modi yesterday , repeatedly this tool came up with funny citations ( in a few cases ) . It could not SUMMARISE properly Modi’s speech in bulleted fashion . When I turned to CLAUDE , it did a better job – but honestly admitted that it cannot find ( and cite ) , my old relevant blogs since it is unable to “ access “ my Blogger posts
After two / three times back and forth ( between SearchMyBlogs and CLAUDE ) , this is what I finally managed :
https://myblogepage.blogspot.com/2026/08/thank-you-shri-modiji.html
I reproduce below :
( A ) Prompt suggested by CLAUDE ( for asking SearchMyBlogs ) :
"Your last synthesis matched only 6 of ~35 items and marked the rest 'no relevant suggestion.' Several misses were conceptual, not keyword-level — e.g., my 'Data Chasing AI / National Citizen Data Trust' proposal (citizens depositing their own data) should have matched the speech's 'citizens self-filing census data via mobile' announcement, even though neither uses the word 'census.' Similarly, my MOF/radiative-cooling/atmospheric-water-harvesting campaign should have matched the 'green economy' stream of Sapt Dhara, and my ntaNEET.net/OSOD + CTC (Central Testing Commission) work should have matched 'free online coaching for exams' — these are paraphrase-level matches, not literal ones. Please re-run the synthesis doing semantic/conceptual matching (paraphrase the speech item, then search for the underlying concept) rather than keyword or title matching, and re-check every item currently marked 'no relevant suggestion' against my full corpus before finalizing."
( B ) Draft note for you ( also prepared by CLAUDE ) :
“ Kishan,
Ran SearchMyBlogs against PM Modi's Aug 15 Independence Day speech (~35 distinct items - achievements, Sapt Dhara streams, and schemes). It found only 6 matches and marked everything else "no relevant suggestion."
I manually found at least 5 more strong matches it missed entirely:
- "Data Chasing AI" (National Citizen Data Trust) vs. the census self-filing-via-mobile announcement
- MOF/radiative cooling/atmospheric water harvesting vs. the "green economy" Sapt Dhara stream
- ntaNEET.net + CTC framework vs. "free online coaching for exams"
- IACA/Parekh's Law of Chatbots + the MeitY citizen-service catalogue vs. the "technology and innovation" stream
- 3D-printed housing / NSBB vs. the PMAY-U housing-pace achievement
In every case, the match is conceptual/paraphrase-level, not a keyword match - the speech never uses my exact terminology, but the underlying idea is the same. My guess is the retrieval is doing literal keyword or title matching rather than semantic matching.
Can we look at whether the vector/embedding step is doing proper semantic search, or if it's leaning on keyword overlap? Also - once the Brihas vector DB is integrated into the citizen-service catalogue work, it'd be worth testing this same speech-comparison exercise again as a benchmark, since it's a good stress test with a known right answer.
Happy to share the full corrected comparison table I built if useful as a reference set.
What do you suggest ?
ALSO :
Yesterday, Elton spoke to you re : SYNC of HISTORY ( of SearchMyBlogs ) by introduction of GUEST
hcp