Sumit Saini @AI__anonymous

Tweets

279
Followers

113
Following

296
Likes

138

Sumit Saini @AI__anonymous

a day ago

Me pretending to be the CEO in an empty office 🤡 and managing my loyal 🪲

0 0 3 43 0

Download Image

Sumit Saini @AI__anonymous

2 days ago

Cooking something on weekends.. will post here soon..

0 0 2 45 0

Sumit Saini @AI__anonymous

3 days ago

Why does my brain force me to post shit on X whenever I see everyone else posting about themselves. 😭

1 0 0 52 0

🚀Introducing Voice + Chat AI Agents on WhatsApp with Sarvam Samvaad Now create rich customer interactions within one seamless thread that includes voice calling + real-time chat. This is powered by: ▪️11 Indian languages across text, voice and widgets ▪️Context-aware,…

31 45 394 24K 104

Download Video

Sumit Saini @AI__anonymous

5 days ago

Voice AI ROI = math, not vibes. >Formula: 80% automation × 50% cost cut × p95 latency <800ms → CSAT↑ cost↓ revenue↑ >Track weekly: autopct handled, AHT, repeat-contact, conv rate. Publish p95s → optimize. #AIROI #EnterpriseAI

0 0 1 40 0

Sumit Saini @AI__anonymous

6 days ago

Great AI knows when to stop. >HITL triggers in Voice AI: >Low ASR confidence >Policy-blocked intents User frustration / repeats >Pass convo state + suggestions to agent → cuts AHT. >HITL isn’t failure, it’s experience insurance. #VoiceAI #nlp

0 0 1 60 0

Sumit Saini @AI__anonymous

7 days ago

Generally the voice bots answer Qs but don’t sell. >My sales agent flow: Qualify → Capture → Recommend → Schedule → Async follow-up → CRM tasks. >Track: lead recycling + multi-product uptake. >Agents should sell like your best rep, daily. #VoiceAI #RevOps

0 0 0 56 0

Sumit Saini @AI__anonymous

a week ago

New month, new grind next 30 days goals: >3 AI/ML + backend projects in voice/conversational systems that solve real business problems >Daily DSA >Open-source contributions in Voice AI >Document learnings

0 0 2 78 0

Sumit Saini @AI__anonymous

a week ago

Today I took the Bangalore Yellow Line metro for the very first time after my exam. sach mein, the crowd was so much that people’s faces were literally touching each other 😭.

0 0 2 137 0

Sumit Saini @AI__anonymous

2 weeks ago

What are the best ways to showcase our skills/projects on X?

0 0 1 60 0

Sumit Saini @AI__anonymous

2 weeks ago

“Hello / नमस्ते / வணக்கம்” — same agent, same quality. Multilingual ≠ translation. Auto language-ID Right ASR/TTS per locale Domain lexicon injection Bilingual fallback when confidence dips Respect local speech = real trust. #IndicAI #VoiceAI

0 0 0 66 0

Sumit Saini @AI__anonymous

2 weeks ago

Most AI agents fail quietly. Hallucinations, robotic tone, poor resolutions. My fix = 100% automated QA: Rubric scoring (facts, tone, policy, resolution) LLM-judge + golden refs Auto-replay fails in sandbox QA isn’t optional. It’s the ROI loop. #AI #VoiceAI #NLP

2 0 7 115 0

Sumit Saini @AI__anonymous

2 weeks ago

Compliance ≠ a sticker. In Voice AI, it’s a pipeline: Regex+NER+phonetic = PII redaction Role-segregated logs TTL on transcripts Consent flags + regional storage HITL for risky actions Secure by design > secure by document. #AI #Security

0 0 2 58 0

Sumit Saini @AI__anonymous

2 weeks ago

How to turn a typo into a holiday 🏖️🤣

0 0 4 84 0

Download Image

Sumit Saini @AI__anonymous

2 weeks ago

AI that only “talks” is half-baked. True power = Conversation → Action. Pattern I use: Intent → ActionResolver → Connector (CRM/ERP/CCaaS). Add retries + DLQ for reliability. Bots that close loops > bots that just chat. #VoiceAI #NLP #SystemsDesign

0 0 1 58 0

Sumit Saini @AI__anonymous

2 weeks ago

Me after annotating antenna on cell towers: Image 69/2000 - already questioning life 😭

0 0 0 46 0

Download Image

Sumit Saini @AI__anonymous

2 weeks ago

15Y/O me: when I wear a T-shirt in Shop Shopkeeper be like: Herooo, Herooo 😭

1 0 2 76 0

Download Gif

Sumit Saini @AI__anonymous

2 weeks ago

If you can’t measure, you can’t improve. In Voice AI, log per-turn events: 1)latency_asr, latency_policy, latency_tts intents, handoffs, sentiment Roll up p95s + win rates → spot drift early & retrain. Analytics ≠ dashboard. It’s a feedback loop. #MLOps #NLP