Nik @capminik

In the Lab Joined November 2014

Tweets

110
Followers

162
Following

2K
Likes

668

Nik @capminik

a month ago

@zach_yadegari I am carrying my phone around with me anyways, why wouldn't I just talk to my ai assistant through my phone. Also this compete for real estate with fitness trackers

0 0 0 230 0

View Details

Nik @capminik

3 months ago

@nickscamara_ @firecrawl would be cool to benchmark it using benchmark released by @llama_index

1 0 0 2K 0

View Details

Nik @capminik

5 months ago

@FactoryAI @rabois interesting

1 0 1 88 0

View Details

Nik @capminik

6 months ago

@trq212 html is deff better for complex tabular data, dealing with merged cells and other edge cases

0 0 0 538 0

View Details

Nik @capminik

6 months ago

@jakobcounts123 @mogulinfluence claude co work is good with pdfs like this

0 1 4 587 2

View Details

@jerryjliu0 Pair the confidence scores with reconciliation on bank statements. Sum extracted transactions against total_deposits and total_withdrawals to confirm every row was extracted. Confidence covers correctness, reconciliation covers completeness.

0 0 0 16 0

View Details

Nik @capminik

6 months ago

@jerryjliu0 trains are flooded with them

0 0 0 52 0

View Details

Nik @capminik

7 months ago

@arcee_ai does it support structured json responses?

0 0 0 67 0

View Details

Nik @capminik

8 months ago

@jerryjliu0 @llama_index Tested this on complex tabular documents and it performs well. Cross page table merging is highly practical for maintaining context when passing to LLM

0 0 0 139 0

View Details

Nik @capminik

8 months ago

@elvissun @openclaw too much friction if you have to verify the oauth each time

0 0 0 302 0

View Details

Nik @capminik

8 months ago

@techNmak I have some OCR docs that fail most models, will test it out

0 0 0 104 0

View Details

Nik @capminik

8 months ago

@SamuelRome97104 @supabase Is there a way to limit to only read permission to avoid destructive queries?

0 0 0 106 0

View Details

Nik @capminik

8 months ago

@real_rehaan @WisprFlow here’s a better example: I just ran into this, “R2” buckets in normal dictation w/o dictionary gets outputed as “are two” When I update “are two” output to “R2” via my keyboard whisper flow auto adds it to my dictionary

1 0 1 22 0

View Details

Nik @capminik

8 months ago

@real_rehaan @WisprFlow an example is I have added the work CLAUDE.md bc I am frequently updating it, initially it would dictate it as two separate words. This is not the best example but think about specific file names in your repos