Sathya (@sathya_ui_dev) header
Sathya (@sathya_ui_dev) profile picture

AI Architect building with GenAI & React ⚙️ Decoding tech, AI & Indian markets in plain English 📈 Chennai | Not investment advice

Chennai
Joined May 2021
Photo by Sathya (@sathya_ui_dev)Photo by Sathya (@sathya_ui_dev)Photo by Sathya (@sathya_ui_dev)Photo by Sathya (@sathya_ui_dev)Photo by Sathya (@sathya_ui_dev)
I'm an AI Architect who builds GenAI products for a living. Every day I break down: 📈 What's moving Indian markets & why 🤖 AI/tech news, from someone who actually builds with it 🛠️ Lessons from my own builds No tips, no hype. Just clear analysis. Follow if that's your thing 👇
4
18
975
🏏 India won the NZ ODI series 3-1. But should Rohit & Kohli still be in the 2027 World Cup plans? Their numbers this tour: • Rohit: 0, 23, 0, 7 = 30 runs, avg 7.50 • Kohli: 74, 7, 0, 5 = 86 runs • 4th ODI: India all out for 89, lost by 244 • 5th ODI: India 209, won by 22. Arshdeep 3/30, Axar & Ishan Kishan 46 each Case to keep them: • Seaming NZ pitches. Shreyas Iyer himself said the series favoured bowlers • Experience counts with the World Cup about a year away Case to move on: • The series was won by the bowlers and middle order, not the top two • Younger openers need ODI reps before 2027 Keep RO-KO for 2027, or build around the next gen? Reply with your pick 👇 #INDvNZ
1
1
40
🎬 Ramayana vs Hollywood CGI: my honest verdict Where it matches Hollywood: • VFX by DNEG, the studio behind Inception, Interstellar and Dune • Animals and world-building look genuinely authentic • The Indra war sequence is Hollywood-standard • Jatayu finally fixes the "plastic creature" problem from Adipurush Where it falls short: • The trailers looked glossy and flat • Quality is uneven across the film, unlike Dune's consistency My take: in its best scenes, Ramayana is on par with Hollywood. Across the full film, not yet. But the debate has moved from "is it better than Adipurush?" to "is it as good as Lord of the Rings?" That's the real win for Indian cinema. Did the CGI work for you? 👇
2
262
💻 Tech update | 6:45 PM IST • Nvidia: Jensen Huang resigned last week and CFO Colette Kress is interim CEO. FT reports Huang is weighing a return, while the US Treasury Secretary wants an outside CEO • OpenAI launches ChatGPT Wallet for AI-agent purchases • White House plans to name 20+ AI firms, including Nvidia, OpenAI and Anthropic, as "national champions" • SpaceX catches a Starship upper stage on the tower for the first time • ULA's Vulcan fails, losing 40 Amazon Leo satellites • GTA VI releases in 3 days My take: AI agents that can spend money is the shift to watch. Once agents can buy things, approvals and spending limits become the real product. #TechNews #AI
1
2
192
10 system design concepts every developer should know before their next interview 👇 1. Load balancing 2. Caching (and cache invalidation) 3. Database indexing 4. Sharding vs replication 5. CAP theorem 6. Message queues 7. Rate limiting 8. Idempotency 9. Consistent hashing 10. Observability: logs, metrics, traces You don't need to master all 10. You need to explain when to use each one, and what it costs you. Which one confuses you the most? #SystemDesign #SoftwareEngineering
17
AI Money Meter 🧮 Amazon bought Whole Foods for about $13.7B in 2017. Big Tech spends that on capex in about 7 DAYS. Amazon bought a grocery chain. Capex bought the whole aisle 🛒 Cart or basket? 👇
6
1
13
345
Only Chennai techies will understand 😂 • OMR traffic decides your login time, not your manager • The filter coffee break is the real standup • Leaving office at 6, reaching home at 8 • "Light rain" in the forecast = WFH request season • Every weekend plan: mall or beach, no third option • Asking "saaptiya?" before asking about the bug Which one is your daily life? 👇 #Chennai #ChennaiRains
100
One line at the top of your system prompt can quietly kill your LLM cost savings 👇 "Current time: {timestamp}" Prompt caching needs an EXACT prefix match. Change one character early in the prompt and everything after it is a cache miss. Why it matters: • Anthropic: cache reads at 0.1x base input price (lower on some models), writes 1.25x • OpenAI: cached input discounted up to 95% What I do now: 1. Static first: instructions, tool definitions, few-shot examples 2. Shared reference/RAG context that repeats next 3. Dynamic last: timestamp, user details, the actual question 4. Append chat turns, never rewrite old ones 5. Don't change tool definitions mid-session, it invalidates the whole cache 6. Log cached tokens per request. If it's ~0, your prompt order is wrong Cheapest optimisation I know. No model switch, just reordering. #GenAI #LLM
35
Every Indian techie has heard these at least once 😂 • "Just a small change, should be quick" • "Can you join the call for 5 mins?" (1 hour later...) • "But it worked in UAT" • "Please do the needful" • "Let's take this offline" • "Hike will be good this year" • "Are you free this weekend? Small production issue" Which one hurts the most? 👇
1
62
Same AI coding tool. Two developers. One ships faster every week, the other gets frustrated and goes back to doing it by hand. The difference is usually 5 habits 👇 1. They give context, not commands. Files, constraints and the goal, not just "fix this". 2. They ask for a plan before the code. 3. They work in small steps and review every diff. 4. They write tests first, so the AI has a clear target. 5. They know when to stop prompting and just read the code. AI doesn't replace engineering judgment. It multiplies whatever judgment you bring. Which habit do you skip most?
6
5
275
Frontend devs, your skills matter more in AI than you think 👇 • State management → handling streaming, partial and failed responses • Loading states → designing for 3–10 second AI waits • Form validation → structured inputs that prevent bad prompts • Accessibility → making AI output readable for everyone • Error handling → turning model failures into useful messages • Design systems → consistent UI for unpredictable output The model is just a backend call. The experience is your job. Frontend folks, what's the first AI feature you'd build? #Frontend #AI
22
📈 Closing bell | 16 Nov Sensex +512 at 82,105 Nifty +147 at 25,291 The day in one line: a strong open, a midday dip, then a recovery into the close. What led: • FMCG and banks • Max Healthcare the top Nifty gainer • FII inflows and softer crude kept the mood positive What didn't: • Midcap 100 -0.5%, Smallcap 100 -1.1% My read: the large-cap vs small-cap gap I flagged at midday got wider. Big names are getting the foreign money while broader stocks see profit booking. That's a selective market, not a broad bull run. Not investment advice. Large caps or small caps for the next 3 months?
1
1
102
Unpopular opinion: "prompt engineer" is not a long-term career. Prompting is becoming a basic skill, like Googling or using Excel. What will actually last: • Deep domain knowledge • Designing evals that catch failures • Building the system around the model: data, tools, guardrails, UX Models keep getting better at understanding what you mean. Your edge is knowing what good output looks like. Agree or disagree?
2
1
110
🎬 Suriya's 2026 is shaping up to be one of his best years. 3 releases, 3 successes: • Karuppu (May): ₹247 Cr worldwide in 11 days • Vishwanath and Sons: ₹31 Cr+ worldwide opening • Scene (Nov): ₹242 Cr worldwide in 10 days Scene is tracking almost level with Karuppu, even while releasing a day after Ramayana. My take: the variety is the real story. A fantasy courtroom drama and an action-comedy, both working at this scale. That's an actor picking scripts, not just a star picking scale. Which one was your favourite?
27
If I were hiring an AI engineer today, these are the 7 questions I'd ask 👇 1. How would you know your RAG system got worse after a change? 2. A user says the bot "made something up". How do you debug it? 3. When would you NOT use an LLM for a feature? 4. How do you cut cost by 50% without hurting quality? 5. What do you log in production, and what do you never log? 6. How do you stop an agent from taking a risky action? 7. Show me something you built that broke. What did you learn? None of these are about which model is best. They're all about judgment. Which one would you struggle with? #AI #GenAI
2
2
168
அன்புள்ள @TVKPartyHQ Sir, சார், பேசாம இந்த ஆட்சியையே கலைச்சிடுங்க. உங்களால இப்போதைக்கு இந்த DMK, ADMK அரசியல் கூட்டத்தைச் சமாளிக்க முடியாதுன்னுதான் தோணுது. எப்பவுமே இந்த DMK, ADMK அரசியல், அவங்களுக்குள்ள நடக்கிற அதிகாரப் போட்டி, பழிவாங்கும் அரசியல் இதையெல்லாம் சமாளிக்கிறதுதான் பெரிய சவாலா இருக்கு. இந்த அரசியல் சூழலைப் புரிஞ்சுக்காம செயல்பட்டா நம்மைச் சுற்றி என்ன நடக்குதுன்னே தெரியாம போயிடும் சார். MLA-க்கள் மேலேயே கேஸ் வருது. உங்களை நம்பித்தான் நாங்க களத்துல இறங்கி மக்கள்கிட்ட ஓட்டு கேட்டோம். ஆனா இப்போ விஷயம் போய்க்கிட்டு இருக்கிற விதம் ரொம்ப சீரியஸா இருக்கு. TVK தரப்புல இருந்து இதுவரைக்கும் ஒரு முறையான அறிக்கையோ, தெளிவான விளக்கமோ வரல. ஒண்ணு மட்டும் புரிஞ்சுக்கோங்க சார். முன்னாள் CM ஸ்டாலினுக்கே கட்சிக்காரங்க தப்பான புள்ளிவிவரங்களைக் கொடுத்து எப்படி ஏமாத்தினாங்களோ, அதே மாதிரி உங்களையும் யாராவது தப்பா வழிநடத்திட்டு இருக்காங்களோன்னு எங்களுக்குச் சந்தேகமா இருக்கு. ஏதோ உள்ளுக்குள்ள சரியில்லாத மாதிரி தெரியுது. Something is fishy! உங்களைச் சுற்றி இருக்கிறவங்க சொல்றதை மட்டுமே நம்பாதீங்க. உண்மையான கள நிலவரத்தைத் தெரிஞ்சுக்க முயற்சி பண்ணுங்க. அரசியல் ரீதியா யார் என்ன செய்றாங்க, எதுக்காகச் செய்றாங்கன்னு புரிஞ்சுக்க வேண்டியது அவசியம். ஒரு பக்கம் போதைப்பொருள் புழக்கம், கொலைகள் எனப் பல பிரச்சினைகள் நடந்துகிட்டு இருக்கு. இன்னொரு பக்கம் TASMAC தொடர்பாகவும் சமூக வலைதளங்கள்ல பல குற்றச்சாட்டுகளும் தகவல்களும் வந்துகிட்டு இருக்கு. இதெல்லாம் உண்மையா, இல்லையா என்பதை விசாரிச்சு TVK தரப்புல இருந்து தெளிவான பதில் கொடுக்கணும். தினமும் வர்ற செய்திகளையெல்லாம் பார்க்கவே முடியல சார். ஒவ்வொரு நாளும் ஏதாவது ஒரு பிரச்சினை, ஏதாவது ஒரு அதிர்ச்சியான செய்தி வந்துகிட்டே இருக்கு. இதையெல்லாம் பார்க்கும்போது நாம எதுக்காக உங்களை நம்பி மக்கள்கிட்ட ஓட்டு கேட்டோம்னு வேதனையா இருக்கு. சார், நாங்க உங்களை நம்பித்தான் மக்கள்கிட்ட போய் ஓட்டு கேட்டோம். உங்களுக்காகக் களத்துல நின்னோம். எங்களோட நம்பிக்கையையும் உணர்ச்சியையும் சாதாரணமா எடுத்துக்காதீங்க. எங்களுக்கு வெறும் பேச்சு வேண்டாம். தெளிவான பதில், வெளிப்படையான செயல்பாடு, உறுதியான தலைமை இதுதான் இப்போ தேவை. நிலைமை கைமீறிப் போறதுக்கு முன்னாடியே உண்மையைப் புரிஞ்சுக்கோங்க சார். எங்களை ஏமாற்றிவிட்டதாக உணர வைக்காதீங்க. நாங்க எதிர்பார்ப்பது ஒரு தெளிவான, பொறுப்பான ஆட்சியை மட்டும்தான். @BussyAnand @CTR_Nirmalkumar @AadhavArjuna
37
🎬 Jailer 2 is already streaming on Prime Video, less than a month after its mid-October release. • ₹826 Cr+ worldwide in theatres, the biggest Tamil film ever • On OTT since 12 Nov, in 5 languages Also new on OTT this week: • Ranabaali (Vijay Deverakonda) on Netflix • Daayra (Kareena Kapoor Khan, Prithviraj) on Netflix My take: a 4-week OTT window is great for viewers but tough on theatres. When fans know even an ₹800 Cr film reaches their phone in a month, many skip the second-week trip. That may be part of why most films drop sharply after week one, while event films like Ramayana in 3D still pull crowds. Theatre or wait for OTT, what's your rule?
85
🎬 #Ramayana is now India's 5th biggest film worldwide. ₹1,384.5 Cr gross in 11 days (Sacnilk) All-time Indian worldwide top 5: 1. Dangal: ₹2,070 Cr 2. Dhurandhar 2: ₹1,813 Cr 3. Baahubali 2: ₹1,788 Cr 4. Pushpa 2: ₹1,742 Cr 5. Ramayana: ₹1,384.5 Cr (still running) Passed on the way: Dhurandhar (₹1,307 Cr), RRR (₹1,230 Cr), KGF 2 (₹1,215 Cr) The quiet story: dubbed versions are holding. Sunday occupancy on far fewer shows: Malayalam 88%, Kannada 66%, Tamil 64% vs Hindi 44% Next stop is Pushpa 2, ₹357 Cr away, with OM: Chapter 1 arriving Friday. Does Ramayana finish in the top 3? 👇 #BoxOffice
153
📈 Markets midday | 16 Nov Sensex +635 at 82,227 Nifty +158 at 25,303 Why it's up: • FIIs keep buying: ₹5,736 Cr net on Friday • Brent crude down ~1% to $95 • IT (+1.1%) and PSU banks (+1.2%) leading What I'm watching: • DIIs have sold for 11 straight sessions. Domestic funds are booking profits while foreign money flows in • Small caps (-0.5%) are lagging. This is a large-cap rally • Nvidia earnings this week could swing IT stocks either way My read: India was the best-performing large market last month (+9.4%, per Geojit). The momentum is real, but a rally this narrow needs broader participation to last. Not investment advice. Are you adding or booking profits?
101
Unpopular opinion: most companies don't need AI agents yet. What they actually need: • Clean, accessible data • Documented processes • One boring automation that saves 10 hours a week Agents amplify whatever system you already have. If your process is a mess, an agent just makes the mess run faster. Fix the workflow first. Add autonomy later. Agree or disagree?
9
8
305
The biggest AI IPO ever is this week. Anthropic, the company behind Claude, in 60 seconds 👇 • Ticker: ANT on Nasdaq • Price range: $70–78 per share • Pricing expected 18 Nov, trading 19 Nov (subject to regulatory clearance) • 825M shares, raising up to ~$64B • ~$1.4T valuation at the top of the range • Plus a $26B private placement with Alphabet, Nvidia, AMD and others • Retail allocation target near 20%, vs the usual 5–10% Why it matters beyond the stock: For the first time, the public market directly prices a frontier AI lab. The result will reset how investors value every AI company, from OpenAI to Indian AI startups. For most Indian investors, access will likely come only after listing, via international investing platforms. Not investment advice. Would you buy on day 1 or wait?
2
3
572