Senior Solution Architect Generative AI - improving search with stats , mathematics , Agents. Bayesian and Evals IS ALL YOU NEED

Woodinville, WA
Released ClinAuthBench v1: a synthetic inpatient health authorization benchmark for testing whether LLMs can reason over dense chart packets without inventing unsupported claims. HF dataset: 📷huggingface.co/datasets/Shiv… GitHub: github.com/Shivi1982/clin-au…
6
16
82,034
Whenever I have low blood pressure i come to google and use their models for max 10-20 minutes they are so dumb they will make your blood boil. Still haven’t met one serious developer who uses Gemini models in production. & it’s a pro model I am using #googleai
6
Shivi Bhatia retweeted
Alaska.
28
407
3,543
38,685
Shivi Bhatia retweeted
President Donald J. Trump officially signs the landmark Greenland security deal alongside Prime Minister Mette Frederiksen of Denmark and Prime Minister Jens-Frederik Nielsen of Greenland. This HISTORIC deal is critical for American national security and defense in the Arctic Region. 🇺🇸🇩🇰🇬🇱
1,574
9,169
45,748
1,403,113
People traveling from #Seattle to any other state via @AlaskaAir is it mandatory for Alaska to always say they are full and need volunteers to move their luggage to check in ??? This is the 5th flight and same ask - is this too low cost of an airline or anything else
1
127
My nephew Michael Hines was born with cerebral palsy. He came into this world with every strike against him, and yet he never had an unhappy day. He was a luminous soul, sent by God to teach us all how to live in gratitude and love. He died a week before his 21st birthday in May 2024. His organs went to nine recipients and saved two lives, including the mother of an intellectually disabled child. His heart went to a nine-year-old boy. After a lifetime of giving the gift of love, he died giving the gift of life. Orlando. April 2024.
2,359
8,039
99,316
1,804,135
Shivi Bhatia retweeted
A 21-year-old man tried to stop an armed robbery during a Belltown street takeover, was shot three times, and walked two blocks begging for help while bystanders recorded him on their phones. No one called 911. KIRO host Gee Scott said this story is exactly why he has no patience for politicians who hide behind spreadsheets and policy buzzwords while families are destroyed. More here: mynorthwest.com/gee-and-ursu…
41
104
557
24,059
Don’t know why it is so hard to work with Claude in an hours time complete pro plan is consumed . And my scripts reads Claude.md first , has rtk token killer for rust garbage, does one script update at a time and no mcp, no dumb skill 1hr TTL caching
1
85
Using meta’s new mule-spark-1.2 and it’s been really really good , the voice mode is damn accurate and the model is really fast , tried glm via open router that model felt like a Stone Age model as compared. @AIatMeta great model
1
37
Shivi Bhatia retweeted
Dad and me feeding Murray... same place, same croc - two photos 15 years apart ❤️🐊
5,231
181,390
989,915
When AI like Claude @claudeai @ClaudeDevs try to become your guardian and cut your chat it’s time to move own. Cancelled my Claude AC and moving to oss either kimi or qwen with open ai largely the main worker . Claude is a commercial platform & they should do that ONLY
2
89
Shivi Bhatia retweeted
High school before smartphones existed
2,356
9,233
150,135
21,741,271
This is open ai 5.6 Sol Ultra - I am not sure if there is any who is building scalable AI LED apps without these issues irrespective of which model. This is frustrating that beyond 200k to and from these models go berserk….
40
Shivi Bhatia retweeted
Sharmila Dhankar. Polio at age 2. Mother blind, father farmer. Married at 18. Beaten for dowry. Beaten for giving birth to 2 daughters. At 26, beaten near fatally, thrown out. At 34, began shotput training. At 40, 1st 🇮🇳 para-athlete 🥇 at CWG.
247
4,413
27,543
359,882
Shivi Bhatia retweeted
Back at the laptop. The usage limits have been reset for all paid users of Codex and ChatGPT Work. Weeeeeeeee. It's a good day!
We’re celebrating the fast adoption of chatGPT Work and all the incredible effort that went into it today. I’m feeling like a limit reset. Hold on tight to your ultra and /fast and see you in a few hours when I’m back at the laptop!
1,560
578
12,922
1,304,803
Not sure how OpenAI works but this is the 7th instance I have this since morning where each instance is around 140k+ context used and it still shows 100% limit left ❤️. This is mind blowing and yes this a pro plan since last 1+ year. @OpenAIDevs 🎊
48
YOu can continue to report best in class benchmarks but in real cases autonomy is too far and model specially claude suck yes you can do a beautiful UI create ppt but real world data like 200K , 100K these models are miserable . this is claude opus 5. @AnthropicAI insanity
62
Shivi Bhatia retweeted
It is really nice to see a crowd in Seattle chanting "USA" rather than burning American flags and marching on behalf of terror organizations Can we import more normal people?
372
2,359
31,170
252,711
Shivi Bhatia retweeted
America is the greatest country in the world and it’s not even close 🇺🇸
7,841
7,181
72,734
11,184,038
Exclusive: The Mercer Island School District, located in the Seattle suburbs, is handing out name change requests for students to change their names/genders without notifying their parents. The fine print reads: "Per Washington State law, if a student requests change(s) listed on this form, the parent/guardian signature is not required."
378
1,574
3,171
349,161
Released ClinAuthBench v1: a synthetic inpatient health authorization benchmark for testing whether LLMs can reason over dense chart packets without inventing unsupported claims. HF dataset: 📷huggingface.co/datasets/Shiv… GitHub: github.com/Shivi1982/clin-au…
6
16
82,034
The differentiator is evidence discipline. Each case includes: dense 72-hour data structured gold label evidence anchors constraints documentation-challenge labels MDP trajectory metadata V1 includes 180 synthetic cases: 108 continued-stay 72 safe/LLOC-ready 22 contradiction/conflicting-documentation cases cases 121-180 include probabilistic MDP traces
1
2
114
No real patient data. No payer/provider/facility/EHR affiliation.this also include a notebook. This is evaluation-scale, not training-from-scratch scale. The goal is to support reproducible work on authorization reasoning, evidence-grounded summarization, contradiction handling, and hallucination control in dense clinical-style documentation
1
55