Sanjay Mittal

Serial entrepreneur, AI technologist. Founder & CEO of Conversational AI Agents co - Predictika. Co-founder Selectica, Spotzot. Focus on AI, ML, NLP.

[Sanjay Mittal @SanjayMittal]

@sabbanms21 @Patriotism87543 Yes. By far. Far worse than James Buchanan, the previous worst. Nixon would be a saint compared to the orange one.

[Sanjay Mittal @SanjayMittal]

It only raised the stakes. And this is not a niche concern: configuration and pricing problems are ubiquitous across the enterprise, whether or not a business calls them that.
I wrote a series of white papers on it. One idea per paper.

[Sanjay Mittal @SanjayMittal]

CPQ has changed on two axes since we pioneered it at Selectica. What we configure grew from a single box to an entire AI data center. How we interact with it went from clicking through a wizard to talking to an AI. The surprising part: neither shift changed the core requirements

[Sanjay Mittal @SanjayMittal]

@Minointerista @MikeZaccardi Alice in Wonderland would be envious!!

[permabera @Minointerista]

@MikeZaccardi it's mostly circular accounting tricks, there are no fucking earnings m8

[Sanjay Mittal @SanjayMittal]

https://t.co/7h4S3Hu9fV

[Sanjay Mittal @SanjayMittal]

Everyone worries their AI will hallucinate facts.
The costlier failure: it has the facts, has the rules, and still applies them wrong.
Logic hallucination. Invisible in a demo, expensive in production
Fix: LLM proposes. Deterministic engine proves.
New CPQ Podcast episode ↓

[Sanjay Mittal @SanjayMittal]

@MelanieJEaston It was a nonsensical movie on so many fronts. On the war front, flying from Leh to Arabian Sea to attack Pak ships must be the biggest joke of all.

[Sanjay Mittal @SanjayMittal]

@MelanieJEaston Maybe she got ChatGPT to write kt for her, though 2200 pages would be a challenge even for a LLM.

[Sanjay Mittal @SanjayMittal]

Final takeaway: The required quality of the AI system is closely tied to the level of autonomy granted to it.
New White Paper: https://t.co/vxdNaA0jhY (https://t.co/vxdNaA0jhY)

[Sanjay Mittal @SanjayMittal]

Closing thought: Did the podcast get everything right? No, it made mistakes, such as hallucinations and difficulty with the term 'RAG.' It also misrepresented the intent of the article in a few instances. However, since the podcast is expository these mistakes can be overlooked.

[Sanjay Mittal @SanjayMittal]

The original white paper discusses the limitations of AI in business reasoning and problem-solving, where correctness and completeness are essential. The podcast makes the contents and arguments from the white paper digestible and explained in a friendly style.

[Sanjay Mittal @SanjayMittal]

There is a fundamental dichotomy between two facets of AI: creative work versus logical consistency. The podcast linked below, showcases AI at its high point by taking a 23-page document and converting it automatically into an interview-style podcast. The result is impressive.

[Sanjay Mittal @SanjayMittal]

• EU261 flight compensation is a perfect stress test — and LLMs fail it consistently
EU Travel Bot: https://t.co/28VFCgPXIZ (https://t.co/28VFCgPXIZ)

Demo Examples: https://t.co/tLmXJ7Q7Oo (https://t.co/tLmXJ7Q7Oo)

[Sanjay Mittal @SanjayMittal]

• Logic-based decisions require deterministic validation, not probabilistic guesses
• Trustworthy enterprise AI requires a hybrid architecture: language + logic
Listen to Predictika podcast: https://t.co/0PUPih6sox (https://t.co/0PUPih6sox)

[Sanjay Mittal @SanjayMittal]

• Prompting, fine-tuning, and RAG improve fluency — not correctness
• Most AI agent stacks let the model grade its own reasoning (a hidden risk)
Listen to Predictika podcast: https://t.co/0PUPih6sox (https://t.co/0PUPih6sox)

[Sanjay Mittal @SanjayMittal]

• LLMs can sound confident and cite sources while still making silent logic errors
Listen to Predictika podcast: https://t.co/0PUPih70e5 (https://t.co/0PUPih70e5)

[Sanjay Mittal @SanjayMittal]

• AI agents are no longer just assistants — they’re making real business and regulatory decisions
Listen to Predictika podcast: https://t.co/0PUPih6sox (https://t.co/0PUPih6sox)

[Sanjay Mittal @SanjayMittal]

https://t.co/RTXsNhN5Go

[Sanjay Mittal @SanjayMittal]

AI agents are moving from chatting to deciding—and that’s where things break. LLMs sound confident but still make silent logic errors that violate rules and regulations. Our new white paper explains why—and how to fix it.
https://t.co/vxdNaA0jhY