@rajisticsRajiv Shah | Explaining AI

Sprinkling AI with Fun 🤣 “criminally underrated” Agent Engineering at OpenHands

card updated Oct 16, 2026 · 12 clips parsed · following 271 · 399,500 likes total

Engagement2.87%below typical

Account summary

@rajistics has 31,300 followers and 1,034 videos. Across the 12 parsed clips the average is 8,679 views and 205 likes. Engagement to views is 2.87%, which is below typical for TikTok. On the newer half of the feed the response is holding steady (−0.17 pp). Posts roughly 0.5 times a week. An average clip reaches 28% of the follower count.

Followers

31.3K

history is still being collected

Engagement

2.87%

−0.17 pp newer half

Average views

8.7K

median 2.1K

Likes per clip

205

12 comments avg

Shares / saves

33 / 52

0.59% of viewers save

Posting rate

0.5 / wk

reach 28% of followers

Latest clips

12 collected

An Obliviate spell for an LLM? NULLs trains models with source-specifi — @rajistics

3.6K1:10An Obliviate spell for an LLM? NULLs trains models with source-specific activation patterns, then excludes a source by s1277230

Was Plato onto something? “Shared Geometry as a Rosetta Stone: Cross-M — @rajistics

1.2K1:19Was Plato onto something? “Shared Geometry as a Rosetta Stone: Cross-Modal Alignment Without Paired Data” shows how sepa664619

Could text diffusion make AI faster? It starts with uncertain token po — @rajistics

2.3K1:16Could text diffusion make AI faster? It starts with uncertain token positions and refines a block of text over several p1304627

61.9K1.2K10027684

Need a primer on retrieval methods? Follow CupidBot as we show when to — @rajistics

1.1K1:10Need a primer on retrieval methods? Follow CupidBot as we show when to use Find, BM25, embeddings and RAG, rerankers, Gr420420

Making an AI agent forget sounds like a bad idea. But on long coding t — @rajistics

2K1:28Making an AI agent forget sounds like a bad idea. But on long coding tasks, it keeps carrying old tool output into every944742

I expected stronger models to need less prompt text. Anthropic’s publi — @rajistics

976I expected stronger models to need less prompt text. Anthropic’s published Opus prompts went from 1,714 words in Opus 4 8002

Most classification problems do not need a chat model writing JSON one — @rajistics

2.6K1:30Most classification problems do not need a chat model writing JSON one token at a time. Jev takes us back to the encoder10711246

DraftKings reportedly built an ML system to predict who would gamble m — @rajistics

1.6K1:15DraftKings reportedly built an ML system to predict who would gamble more after a promotion. 533216

20.6K0:09Reality of AI models 344111936

You want big picture thinking about AI Projects. Check out all resourc — @rajistics

2K1:29You want big picture thinking about AI Projects. Check out all resources for AI Problem Framing on my maven site over at11722492

Coding agents aren’t just models. The harness around the model control — @rajistics

4.4K34:14Coding agents aren’t just models. The harness around the model controls the loop: retrieval, memory, tool use, safety, t2161034205

Engagement per clip

12 parsed clips · newest on the right · click a bar to open the clip peak: 7.2% ER

May 3Sep 27Oct 11

Date	Views	Likes	Comments	Shares	Saves	ER
Oct 11, 2026	3,569	127	7	2	30	3.81%
Oct 10, 2026	1,155	66	4	6	19	6.58%
Oct 4, 2026	2,264	130	4	6	27	6.18%
Oct 2, 2026	61,900	1,151	100	276	84	2.47%
Sep 28, 2026	1,092	42	0	4	20	4.21%
Sep 27, 2026	1,980	94	4	7	42	5.30%
Sep 23, 2026	976	8	0	0	2	0.82%
Sep 20, 2026	2,620	107	1	12	46	4.58%
Sep 19, 2026	1,569	53	3	2	16	3.70%
Sep 18, 2026	20,600	344	11	19	36	1.82%
May 17, 2026	1,986	117	2	24	92	7.20%
May 3, 2026	4,432	216	10	34	205	5.87%