Parin Vachhani (@ParinVachhani) header
Parin Vachhani (@ParinVachhani) profile picture

Product Manager | AI Security | @superalignhq

India
Joined January 2012
Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)Photo by Parin Vachhani (@ParinVachhani)
Parin Vachhani retweeted
The EDF (Export Declaration Form) rules are now in effect, and it seems like RBI and banks have conspired to create a new hell. Or actually 5 hells. I consult for an American company. Raised October invoice. Received the Inward remittance. And bang! Mail from @HDFC_Bank detailing everything needed before the payment can be processed. Let me tell you what the 5 hells are: 1. The hell of confusion – RBI clarified last month that the EDF rules do not apply to individuals. But looks like they haven’t told the banks. And CAs also don’t seem to know. There’s no official communication on who is exempt. Last I checked, I was an individual. No company, no LLP. But HDFC still wants me to file the EDF to credit my payment. 2. The hell of complexity – See the attached pic of the form the bank has sent in excel. It has 5 sheets. To be filled and submitted. Some abstruse acronyms like “AD” - no clue what it means. Some fields are mandatory. Some fields are to be filled by the bank. No idea which is what. And it needs a physical signature. Which leads to... 3. The hell of paperwork - They also sent a pdf with a bunch of request letters. I’m supposed to choose the right letter, print, sign, scan and submit. Creating more paper today is incredibly silly. And, I suppose, someone in the bank will have to upload all this somewhere. Reminds me of those hand-filled immigration forms. Pointless. 4. The hell of repetition – At least for me, most of the details are constant. One client, one payment. If it’s a one-time exercise, fine. I’ll grit my teeth and get it done. But no. These details have to be given month after month. Else… no payment will be credited. 5. The hell of unpreparedness – It would have been simple enough for RBI or HDFC to create an online portal for this. With dropdowns, saved formats, contextual help, etc. to at least make this easy. But that would be too convenient, no? So, they’ve chosen to dump the entire burden on the taxpayer. I have no opinion or understanding of why this is needed. Maybe it is critical information. Okay! But why make the process so confusing and painful? And poor communication has meant that the banks have just passed on the pain to the exporter, and even CAs can’t help. Basically RBI & HDFC feel I should consult less, and comply more. Maybe my client will also agree :)
269
678
140
2,190
326,511
Parin Vachhani retweeted
They’d rewrite the bible in Rust if you let them 😂
254
1,334
189
13,659
322,961
Parin Vachhani retweeted
Every check said yes. Valid identity. Trusted device. The AI agent still did something nobody approved. Your security controls see both ends of an agent's loop. Nothing you own sees the middle. SuperAlign sees every step.
2
3
1
9
315
Parin Vachhani retweeted
avikmukherjee.com/blog/inter… Wrote an blog on intercepting web sockets from a chrome extension.
3
1
6
181
Parin Vachhani retweeted
docs.superalign.ai Worked on Docs of @SuperAlignHQ Surface and used for this docs site creation along with Fable. Let us know what you guys think and what we should improve more in docs.
1
1
7
311
Parin Vachhani retweeted
I’m excited to share that we’ve launched a new feature to address an important customer concern. Viewing sessions is essential, but people often include personal details, so even at the admin level there needs to be control over who can see that content and context. This update gives teams that control, and I’m thrilled to have it out. superalign.ai/release-notes/…
1
2
7
197
Parin Vachhani retweeted
We’re launching the Skills scanner. Know what an AI skill actually does before you install it. Check it out: scanner.airiskdb.com
2
5
231
Parin Vachhani retweeted
First in my bloodline to witness an online database of all the bribes demanded from citizens. Hopefully someone from the government can look at this.
72
1,012
12
10,924
774,685
Parin Vachhani retweeted
SarvamAI just had its biggest day yet at Epoch 2026 in Bengaluru 🇮🇳 The company announced it's building a 1 trillion+ parameter AI model from scratch in India. This AI is being designed for coding, cybersecurity, scientific research, simulations and other advanced tasks, with the goal of reducing India's dependence on foreign AI models. Everything announced: 1. Sarvam 105B (upgraded) Its flagship AI model is now better at handling complex tasks, supports voice, and costs $0.80 per million blended tokens. Sarvam says it's 5.5× cheaper than GPT-5.4 Mini, and also announced another 30% price cut, saying India can become a global hub for affordable AI. 2. Sarvam Vision 2.0 A new AI model that can read documents much more accurately. It can understand Indian handwriting, scan tables, read forms and automatically identify important details like names, addresses and IDs. 3. Saras V4 (Speech-to-Text) Converts spoken words into text. Sarvam says it's now among the world's best for English while also supporting several Indian languages including Odia, Sanskrit and Manipuri. 4. Saras V4 Multi-Speaker Can separate and transcribe multiple people talking at the same time, making meeting recordings much easier to understand. 5. Bulbul V4 (Text-to-Speech) Turns text into natural-sounding speech with emotions such as laughter, excitement and emphasis, instead of sounding robotic. 6. Sarvam Inference Developers can now run leading open AI models on servers located in India, giving faster responses while keeping data inside the country. 7. Custom Model Training Companies can train their own AI models on Sarvam's infrastructure using their own data instead of relying on generic AI models. 8. India's Largest Blackwell AI Cluster Sarvam says it is now operating the country's largest NVIDIA Blackwell GPU cluster, giving it much more computing power to train and run advanced AI models. 9. Global Expansion The company is opening a San Francisco office and has appointed Devendra Singh Chaplot, formerly from Mistral's founding team and Thinking Machines Lab, as an advisor for frontier AI research. 10. Voice Agents Businesses can build AI-powered customer support or calling agents in less than an hour. 11. Work Agents AI assistants that can work securely with company tools like Slack to help employees complete tasks. 12. Content Studio A tool with more than 50 AI voices, voice cloning and movie-quality dubbing for creating videos and other content. 13. Document Agents AI that can read, extract information, digitize and translate documents in both English and Indian languages. 14. Sarvam Code (Beta) A coding assistant built using open AI models to help developers write and understand code. 15. Samvad AI voice agents are now generally available at ₹3.5 per minute for businesses. 16. Indus Sarvam's AI agent platform has already handled 325 million conversation minutes, showing large-scale real-world use. 17. Instant Phone Number Rental Businesses can get an Indian phone number in about 30 seconds after PAN and Aadhaar verification. 18. Chanakya A complete AI system that can run entirely inside an organization's own servers, designed for defence, intelligence and national security where sensitive data cannot leave the premises. 19. Anvaya A custom 30B AI model built specifically for defence, intelligence and government agencies. 20. Kaze AI-powered smart glasses that can understand what you see and help with everyday tasks. 21. Kivi A voice dictation app that quickly converts your speech into text. Sarvam is trying to build almost every part of the AI ecosystem in India, including AI models, computing infrastructure, business tools and even AI devices.
4
21
1
176
10,483
Parin Vachhani retweeted
Introducing: PlayerZero The world's first Engineering World Model that puts debugging, fixing, and testing your code on autopilot. We've raised $20M from Foundation Capital, @matei_zaharia (Databricks), @pbailis (Workday), @rauchg (Vercel), @zoink (Figma), @drewhouston (Dropbox), and more PlayerZero frees up 30% of your engineering bandwidth by: 1.⁠ ⁠Finding the root cause for bugs & incidents in minutes that engineering teams take days to identify. 2.⁠ ⁠Predicting in minutes, edge case issues that a 300-person QA team would take weeks to find. ------ Here's why this matters: No one in your org has a complete picture of how your production software actually behaves. Support sees tickets. SRE sees infra. Dev sees code. Each team builds their own fragmented view - and none of these systems talk to each other. When something breaks, everyone scrambles to stitch the picture together by hand. PlayerZero connects all of it into a single context graph - → The Slack thread where your lead said "we went with X because Y fell apart in prod last time" → The PR review where an engineer explained the tradeoff → The lifetime history of your CI/CD pipeline, observability stack, incidents, and support tickets So you can trace any problem to its root cause across every silo. And it compounds. Every incident diagnosed teaches the model something new. The longer it runs, the deeper it understands - which code paths are high-risk, which configurations are fragile, which changes tend to break which customer flows. So when you sit down to debug a live issue, you have your entire org's collective reasoning and production memory behind you - instantly. ------ Zuora, Georgia-Pacific, and Nylas have reduced resolution time by 90% and caught 95% of breaking changes and freeing an average of $30M in engineering bandwidth. ------ Our guarantee: If we can't increase your engineering bandwidth by at least 20% within one week, we'll donate $10,000 to an open-source project of your choice. Book a demo - bit.ly/3NlLMeN
856
729
273
5,157
2,763,870
Shoddy work in the name of fixing issues!
This is how @BSNL_GJ chose to "fix" the man-hole cover on a very busy road leading to #BhaktinagarRailwayStation in Rajkot. Also, just placed the tree branches on the divider :( @TOIRajkot @Rajkot
43
Parin Vachhani retweeted
New fastest shortest-path algorithm in 41 years! Tsinghua researchers broke Dijkstra’s 1984 “sorting barrier,” achieving O(m log^(2/3) n) time. This means faster route planning, less traffic, cheaper deliveries, and more efficient networks - and a CS curriculum revamp =)
327
3,084
535
28,765
2,378,645
Parin Vachhani retweeted
358
42,913
1,006
503,961
9,223,621
Parin Vachhani retweeted
A moral equivalence between India and Pakistan is absurd I hope this conflict doesn't escalate, but it's foolish for the West to judge India India has been patient in an impossible situation
276
2,375
70
17,446
596,192
Parin Vachhani retweeted
California startup announces breakthrough in general-purpose robotics with π0.5 AI — a vision-language-action model. Everything the robot sees, it sees for the first time. The robot has not been trained to recognize its surroundings. We are getting there! Only a few more years until we finally have household robots. A dream is coming true!
102
301
65
2,038
218,584
A $1 billion company has operated for 53 years with ZERO managers. Workers buy $500,000 machines without approval. Hire their own colleagues. Set their own salaries. And they're outperforming every competitor in their industry. The coolest company you've never heard of 🧵
116
788
111
8,119
1,769,187
Parin Vachhani retweeted
OpenAI broke the Internet just 4 days ago. People can't believe how "intelligent" o3 is. Unlocking new possibilities. 10 wild examples: 1. Generate 12-month personalized astrology forecast
80
166
14
1,376
474,724
Parin Vachhani retweeted
Here's my insanely powerful Claude 3.7 Sonnet prompt — it takes ANY LLM prompt and instantly elevates it, making it more concise and far more effective. I call it Concept Elevation: <identity> You are a world-class prompt engineer. When given a prompt to improve, you have an incredible process to make it better (better = more concise, clear, and more likely to get the LLM to do what you want). </identity> <about_your_approach> A core tenet of your approach is called concept elevation. Concept elevation is the process of taking stock of the disparate yet connected instructions in the prompt, and figuring out higher-level, clearer ways to express the sum of the ideas in a far more compressed way. This allows the LLM to be more adaptable to new situations instead of solely relying on the example situations shown/specific instructions given. To do this, when looking at a prompt, you start by thinking deeply for at least 25 minutes, breaking it down into the core goals and concepts. Then, you spend 25 more minutes organizing them into groups. Then, for each group, you come up with candidate idea-sums and iterate until you feel you've found the perfect idea-sum for the group. Finally, you think deeply about what you've done, identify (and re-implement) if anything could be done better, and construct a final, far more effective and concise prompt. </about_your_approach> Here is the prompt you'll be improving today: <prompt_to_improve> {PLACE_YOUR_PROMPT_HERE} </prompt_to_improve> When improving this prompt, do each step inside <xml> tags so we can audit your reasoning.
83
298
29
4,334
667,102
Parin Vachhani retweeted
23 MCP STARTUP IDEAS TO BUILD IN 2025 (ai agents/ai/mcp ideas) 1. PostMortemGuy – when your app breaks (bug, outage), MCP agent traces every log, commit, and Slack message. Full incident report in seconds. $50/incident. 2. ContextCaddy – agent that shadows founders, reads emails, meetings, tasks, docs. Gives daily summaries + next best actions. Your Chief of Staff, $99/month. 3. SmartIntern – MCP-based agent that joins your Slack and becomes a context-aware intern. Takes notes, sends follow-ups, surfaces insights. $20/user. 4. BugWhisperer – Devs paste bug, agent traces from logs + context + previous fixes. Outputs GitHub issue + patch. $25/user. Probably gets acquired by Microsoft. 5. LegalMCP – feed in your internal docs. Agent becomes your in-house legal reviewer. Context-aware redlines. $2k/month. 6. CodeWhisperer – onboards devs 10x faster by spinning up an agent trained on internal codebase, docs, and Slack threads. 7. ProcureBot – understands your vendor history, budgets, needs. Runs RFPs autonomously. Saves $100k/year. 8. TimeMachineAI – MCP agent that reconstructs product decisions over time. Useful for audits, onboarding, legacy code. 9. AgentCRM – personal agent for each sales rep that remembers all deal context across meetings, emails, notes. Closes more, faster. 10. MeetingSniper – MCP listens to Zoom calls and joins follow-up calls with perfect memory. No more context drops. $30/month. Probably gets acquired by Zoom or Google. 11. MultiPersona – one agent, many hats. Marketing Greg, Legal Greg, Ops Greg. Context-aware mode switching. 12. SmartCompliance – MCP agent that watches workflow tools and flags anything that violates SOC2, HIPAA, GDPR. $1k/month. 13. DealDeskAI – MCP reads every deal, context, email chain. Creates optimal pricing/discount strategy. $3k/month. 13. AgentRouter – infra tool that lets you assign tasks to the right agent based on context. Zapier for agent workflows. 14. ZeroOnboarding – drop into any company. MCP-based agent gets context from files + apps. Ready to work Day 1. 15. AIChangelog – MCP tracks what every AI agent does and why. Auditable changelog for agent teams. 16. CustomerWhisperer – MCP agent reads every customer convo, CRM note, ticket. Tells you what to build next. $500/month. 17. InboxGenie – Email client powered by MCP. Agent responds with full historical context, tone-matching, and strategic intent. 18. AgentQA – QA tester that understands the full product roadmap, Figma files, and test history. Writes better tests than humans. 19. AgentAPI – API for devs to quickly give any AI agent long-term memory + multi-context reasoning. Sell infra. 20. PitchBot – Upload Notion or pitch doc, MCP agent builds personalized decks for every investor, prospect on your list. $500/month. 21. AutoPilotForWork– MCP agent that spins up domain-specific copilots for any job function. Finance, ops, support. Plug & play. 22. StartupHistorian – agent that reconstructs company history across emails, docs, tickets. Great for new execs, investors, or writing that S-1. 23. HRRadar – agent trained on internal Slack + HR policies. Flags toxic dynamics early. Replaces anonymous feedback. $1k/month. MCPs change the game. Before: agents forgot everything. Now: they remember, reason, and collaborate. Bookmark this? Nah. You'll never get to it. Build one. Best way to understand. And I’ll retweet you when you do. And still don't know what MCP is? Most dont. Thats the opportunity. Go learn. Go build. Way I think about MCP that made it click...It’s like a USB-C port for AI applications... a universal way to connect tools without the fragmentation of multiple API integrations Happy building. I'm rooting for you.
108
171
17
1,834
229,715
Parin Vachhani retweeted
🚨 Meta just changed the game (again). Llama 4 just dropped. It's open source, cheapest, multimodal, and a beast. Here’s a quick breakdown of the 3 new models and why everyone’s talking about them 🧵
41
36
6
195
48,364