DEV Community

Ashutosh Singhal profile picture

Ashutosh Singhal

Founder, Veriprajna | Building True Intelligence • LinkedIn: https://www.linkedin.com/in/ashutosh-veriprajna/

The CrowdStrike crash came down to a field count: 21 where the kernel expected 20. No independent layer was checking.

The CrowdStrike crash came down to a field count: 21 where the kernel expected 20. No independent layer was checking.

Comments
9 min read

Want to connect with Ashutosh Singhal?

Create an account to connect with Ashutosh Singhal. You can also sign in below to proceed if you already have an account.

Already have an account? Sign in
I social-engineered an AI game guard into giving up a key it was built to protect. Its twin wouldn't budge.

I social-engineered an AI game guard into giving up a key it was built to protect. Its twin wouldn't budge.

Comments
9 min read
The Naive Baseline in My Radar Fall Detector Hit the Same 1.0 Recall. It Also Fired Seven False Alarms in One Night.

The Naive Baseline in My Radar Fall Detector Hit the Same 1.0 Recall. It Also Fired Seven False Alarms in One Night.

Comments
11 min read
I Built a Solar Firewall for Curved Glass Facades. Its Most Useful Answer Is Sometimes No Number.

I Built a Solar Firewall for Curved Glass Facades. Its Most Useful Answer Is Sometimes No Number.

Comments
8 min read
I built an AI to recover the black plastic recyclers throw away. The classifier was the easy part.

I built an AI to recover the black plastic recyclers throw away. The classifier was the easy part.

Comments
10 min read
An automated camera followed a bald head for a whole soccer match. I built the physics gate that says no.

An automated camera followed a bald head for a whole soccer match. I built the physics gate that says no.

Comments
8 min read
I fed raw radiance to a hyperspectral crop AI. On a held-out synthetic test, its score fell from 0.887 to 0.373.

I fed raw radiance to a hyperspectral crop AI. On a held-out synthetic test, its score fell from 0.887 to 0.373.

Comments
8 min read
Eight Green Formal Proofs Became Five Fileable Ones When I Audited the SystemVerilog Assertions

Eight Green Formal Proofs Became Five Fileable Ones When I Audited the SystemVerilog Assertions

Comments
8 min read
A satellite said a depot was under water. It was a cloud's shadow, and the parametric trigger was one click from auto-paying...

A satellite said a depot was under water. It was a cloud's shadow, and the parametric trigger was one click from auto-paying...

Comments
10 min read
A perfect model can't stop your self-driving lab from running a doomed synthesis. I built the gate that does.

A perfect model can't stop your self-driving lab from running a doomed synthesis. I built the gate that does.

Comments
10 min read
The same body is a 29 in one jean and a 28 in another. I built the fit engine that explains why a size chart can't tell them...

The same body is a 29 in one jean and a 28 in another. I built the fit engine that explains why a size chart can't tell them...

Comments
10 min read
A label tagged its catalog with C2PA for the EU AI Act. I watched one transcode strip all of it.

A label tagged its catalog with C2PA for the EU AI Act. I watched one transcode strip all of it.

Comments
11 min read
Any pose model will log 15 good squats on a knee that is caving. I built the AI that catches rep 9.

Any pose model will log 15 good squats on a knee that is caving. I built the AI that catches rep 9.

Comments
10 min read
I set out to build a deepfake detector for auto insurance claims. A real, unedited photo beat it.

I set out to build a deepfake detector for auto insurance claims. A real, unedited photo beat it.

Comments
9 min read
My AI drafted a polished cold email. It was also a lie, and I built the gate that stopped the send.

My AI drafted a polished cold email. It was also a lie, and I built the gate that stopped the send.

Comments
10 min read
I built an AI to beat the solver at airline crew recovery. It lost, and that loss became the product.

I built an AI to beat the solver at airline crew recovery. It lost, and that loss became the product.

Comments
9 min read
My knowledge-tracing AI wanted to certify a learner who gamed the course. The code I wrote refused.

My knowledge-tracing AI wanted to certify a learner who gamed the course. The code I wrote refused.

Comments
10 min read
My drone holds 0.9% drift through a GPS jamming bubble. The code I am proud of is the line that refuses to fly.

My drone holds 0.9% drift through a GPS jamming bubble. The code I am proud of is the line that refuses to fly.

Comments
11 min read
My newsroom AI invented a quote the mayor never said. Catching it before readers did is the whole product.

My newsroom AI invented a quote the mayor never said. Catching it before readers did is the whole product.

Comments
12 min read
My booking agent told a traveler "you're all set" with no hotel. I stopped trusting the LLM to run the flow.

My booking agent told a traveler "you're all set" with no hotel. I stopped trusting the LLM to run the flow.

Comments
9 min read
A single COBOL file told the AI everything except the one fact that mattered, so I built the map first.

A single COBOL file told the AI everything except the one fact that mattered, so I built the map first.

Comments
12 min read
I built a demo to reproduce a famous AI mistake. My baseline refused to make it.

I built a demo to reproduce a famous AI mistake. My baseline refused to make it.

Comments
10 min read
The statute was real. The answer was still illegal.

The statute was real. The answer was still illegal.

Comments
12 min read
A model rated a visibly wrong red 0.97 on-brand. That number changed what I built.

A model rated a visibly wrong red 0.97 on-brand. That number changed what I built.

Comments
12 min read
I built a flood model that beat FEMA's map. Then it refused to let me file it.

I built a flood model that beat FEMA's map. Then it refused to let me file it.

Comments
9 min read
I Watched a Private LLM Leak a Board Document. The Model Did Nothing Wrong.

I Watched a Private LLM Leak a Board Document. The Model Did Nothing Wrong.

Comments
12 min read
I Kept Trying to Fix the Model. The Problem Was the Light.

I Kept Trying to Fix the Model. The Problem Was the Light.

Comments 2
12 min read
Three chatbots, zero safety violations, one tribunal ruling. I had the problem labelled wrong.

Three chatbots, zero safety violations, one tribunal ruling. I had the problem labelled wrong.

Comments
13 min read
The claim was true. I killed it anyway.

The claim was true. I killed it anyway.

Comments
12 min read
I read a mental-health chat where no single message was dangerous. That was the danger.

I read a mental-health chat where no single message was dangerous. That was the danger.

Comments
10 min read
The cases were real. I still couldn't let the AI vouch for them.

The cases were real. I still couldn't let the AI vouch for them.

Comments
12 min read
I Tried to Make an LLM Catch Its Own Tax Errors. It Can't, and That Turned Out to Be the Whole Business.

I Tried to Make an LLM Catch Its Own Tax Errors. It Can't, and That Turned Out to Be the Whole Business.

Comments
11 min read
Your CFO's Face Costs $50 to Fake. Your Wire Controls Weren't Built for That.

Your CFO's Face Costs $50 to Fake. Your Wire Controls Weren't Built for That.

Comments
11 min read
Your AI Models Are Executable Code. Most Companies Treat Them Like Spreadsheets.

Your AI Models Are Executable Code. Most Companies Treat Them Like Spreadsheets.

Comments
11 min read
1,500 Megawatts Vanished in 82 Seconds — And the Grid Operators Never Saw It Coming

1,500 Megawatts Vanished in 82 Seconds — And the Grid Operators Never Saw It Coming

Comments
9 min read
73,000 Smart Meters Went Dark Overnight — And the Firmware Had Passed Every Lab Test

73,000 Smart Meters Went Dark Overnight — And the Firmware Had Passed Every Lab Test

Comments
9 min read
We Built a Fake Review Detector That Worked Perfectly — Until Someone Ran It Through BypassGPT

We Built a Fake Review Detector That Worked Perfectly — Until Someone Ran It Through BypassGPT

Comments
10 min read
Your AI Shopping Assistant Converts 4x — Until It Makes One Thing Up

Your AI Shopping Assistant Converts 4x — Until It Makes One Thing Up

Comments
10 min read
18,000 Cups of Water: What Drive-Thru Voice AI Keeps Getting Wrong

18,000 Cups of Water: What Drive-Thru Voice AI Keeps Getting Wrong

Comments
11 min read
418 Kilovolts Looked Fine. Then 60 Million People Lost Power.

418 Kilovolts Looked Fine. Then 60 Million People Lost Power.

Comments
9 min read
The Green Dashboard Lied: What Enterprise AI Validation Actually Has to Test

The Green Dashboard Lied: What Enterprise AI Validation Actually Has to Test

Comments
11 min read
Your Save Flow Is Quietly Manufacturing Churn — Here's the Number That Proved It

Your Save Flow Is Quietly Manufacturing Churn — Here's the Number That Proved It

Comments
10 min read
Your Procurement AI Scored a Better Supplier Lower — Here's the Math That Did It

Your Procurement AI Scored a Better Supplier Lower — Here's the Math That Did It

Comments
9 min read
Apple and Goldman Lost Thousands of Disputes Into a State Nobody Knew Existed. It Cost $89 Million.

Apple and Goldman Lost Thousands of Disputes Into a State Nobody Knew Existed. It Cost $89 Million.

Comments
10 min read
Eleven Words in an Annual Report Can Now Be Securities Fraud

Eleven Words in an Annual Report Can Now Be Securities Fraud

Comments
10 min read
Your Algorithm Sold $1.4 Billion Before Anyone Hit Cancel. Can You Explain Why?

Your Algorithm Sold $1.4 Billion Before Anyone Hit Cancel. Can You Explain Why?

Comments
10 min read
90% of AI Coverage Denials Get Reversed on Appeal. A Court Just Called That a Breach of Contract.

90% of AI Coverage Denials Get Reversed on Appeal. A Court Just Called That a Breach of Contract.

Comments
10 min read
A Court Ruled Your Chatbot Is a "Product." That One Word Rewrote AI Product Liability.

A Court Ruled Your Chatbot Is a "Product." That One Word Rewrote AI Product Liability.

Comments
11 min read
Ninety Percent of Physicians Trust Their Clinical AI. They Catch a Third of Its Dangerous Errors.

Ninety Percent of Physicians Trust Their Clinical AI. They Catch a Third of Its Dangerous Errors.

Comments
11 min read
A Facial Recognition Vendor Scored 99% — and Still Flagged Innocent Shoppers All Day

A Facial Recognition Vendor Scored 99% — and Still Flagged Innocent Shoppers All Day

Comments
10 min read
Your Tenant-Screening AI Passed the Fairness Test. It Was Still Discriminating.

Your Tenant-Screening AI Passed the Fairness Test. It Was Still Discriminating.

Comments
11 min read
Your Pricing Algorithm Is Two Lawsuits, Not One — And Most Companies Only Defend Against One

Your Pricing Algorithm Is Two Lawsuits, Not One — And Most Companies Only Defend Against One

Comments
10 min read
We Built One Bias Audit to Satisfy Six AI Hiring Laws. It Failed in the Same Week We Shipped It.

We Built One Bias Audit to Satisfy Six AI Hiring Laws. It Failed in the Same Week We Shipped It.

Comments
10 min read
8.5 Million Computers Crashed From One File Nobody on Your Side Reviewed

8.5 Million Computers Crashed From One File Nobody on Your Side Reviewed

Comments
10 min read
I Watched a Playtester Talk an AI Merchant Out of a Quest Key With One Sentence

I Watched a Playtester Talk an AI Merchant Out of a Quest Key With One Sentence

Comments
12 min read
The Bathroom Breaks Every Fall Detection System — Here's What We Built Instead

The Bathroom Breaks Every Fall Detection System — Here's What We Built Instead

Comments
11 min read
The Render Was Beautiful. The Building Couldn't Be Built.

The Render Was Beautiful. The Building Couldn't Be Built.

Comments
10 min read
Your Optical Sorter Is Blind to Black Plastic. We Stopped Calling It a Software Problem.

Your Optical Sorter Is Blind to Black Plastic. We Stopped Calling It a Software Problem.

Comments
11 min read
The AI Tracked a Bald Head for the Whole Match. More Training Data Won't Fix That.

The AI Tracked a Bald Head for the Whole Match. More Training Data Won't Fix That.

Comments
10 min read
A Pharma Drug-Discovery Model Designed 40,000 Toxins in Six Hours. The Fix Wasn't What I Expected.

A Pharma Drug-Discovery Model Designed 40,000 Toxins in Six Hours. The Fix Wasn't What I Expected.

Comments
12 min read
loading...