Happy to share that my project was selected as one of the winners in the AI Innovation challenge organized by Google.
This is what I built: drones fly along power lines and capture thermal images plus telemetry like GPS and temperature right in the field. All of that data streams in real time through Confluent Kafka, so there’s no batch-processing lag — everything is live. Google’s Gemini 2.5 Flash then analyzes each image as it comes in, picking up thermal anomalies that simple temperature readings would miss. When a hotspot is detected, it instantly shows up on a live dashboard with a risk score, GPS coordinates, and an AI-generated explanation so operators know what’s going on and what to look at first.
This win gives me the push to keep learning and work on complex problems. Thanks to Google and Confluent for providing the platform!
the exact n8n workflow that's printing viral tweets for me
(and why I'm borderline stupid for sharing this)
most creators are burning 2+ hours daily crafting tweets that flop
meanwhile, I text my Telegram bot and get publication-ready content in 30 seconds
here’s what happens:
→ send message to Telegram bot
→ claude generates hook + full post
→ tweet gets sent automatically
→ ask for iterations if needed
the secret sauce isn't the workflow (that's easy)
It's the prompt engineering behind it
I reverse-engineered how top copywriters structure hooks
and systematized it into one prompt that consistently outputs 6-7 figure level copy
this thing cranks out posts that get:
- 10x more engagement than "normal" tweets - built-in psychological triggers
- perfect formatting for mobile scanning - hooks that stop the scroll dead
most people are still manually writing tweets like it's 2019
smart operators are automating their entire content pipeline
while maintaining quality that puts agencies to shame
the workflow takes 3 minutes to set up
the prompt took me almost 50 iterations to perfect
want the complete system?
Comment "WORKFLOW" + follow + repost and I'll send you everything
the prompt alone is worth more than most $497 courses
@GabbbarSingh Building software apps - most of the code is through ai, spending more time in planning and specs. Basically AI has changed the way we used to developed software, now it’s going towards specs driven dev.
Just submitted my pitch for the @perplexity_ai Computer Stock Pitch Competition!
Long Vertiv Holdings (NYSE: VRT)12–18 month price target: $320–$370 (14–32% upside)
Used Perplexity Computer for researching all the financials and tecnical analysis.
@perplexity_ai@askperplexity@PPLXFinance
#PerplexityComputer
@azuresupport#azhelp:
I am getting this error while creating resource for Azure Foundry Open AI
contact support.. (Code: RequestDisallowedByAzure, Target: proj-default) I am using my student account to use Azure Foundry resourse
Prabhu Shree Ram had the most powerful horoscope ever cast. Five planets exalted or in their own sign Three Mahapurusha Yogas. Jupiter in the lagna.
And he still lost his kingdom. His home.The woman he loved.That is not a contradiction.That is the deepest teaching Vedic astrology has ever produced.
1/15 🧵
Built **Go-live** — a real-time AI Go-live agent that sees your router/breaker/lock via camera, talks naturally, interrupts gracefully, reads manuals you show it, and fixes most issues in minutes. No more waiting for a $100+ pro call.
Powered entirely by Google AI + Cloud. Here’s how we did it. 🛠️ #GoogleAI #GeminiLive #Hackathon
Core brain: **Gemini Live API** (via Google GenAI SDK)
- Real-time bidirectional streaming (voice + ~1 FPS vision)
- Native barge-in/interruptions (user can talk over it)
- Affective dialog (calms down if you sound frustrated)
- Model used: gemini-2.5-flash-native-audio-preview
It literally hears you, sees what you point the camera at, and responds in natural voice. Magic.
Frontend: React + Vite + TailwindCSS (browser/mobile)
- Live camera feed + mic capture
- Cyberpunk-style AR overlays (cyan highlights on WAN port, amber arrows on loose cable)
- WebSocket streaming of PCM audio & JPEG frames to Gemini Live
No heavy native AR frameworks — kept it web-first for fast iteration.
Backend/Deployment: **Google Cloud Run** (containerized)
- Dockerized Node.js server
- Handles session management, API key proxy (never expose key client-side)
- Auto-scales to zero when idle — cheap for hackathon
- Deployed via `gcloud run deploy` — took ~5 minutes
Prompt engineering was the crucial:
- Single ordered system prompt: intent classification → safety check → spatial/vision comments → 5-step flow
- Added manual reading: show page → extracts LED chart/port diagram → cross-references live view
- Strict rules: no internal thoughts, one short sentence max, hallucination prevention
Result: precise, no-fluff responses that feel like a real diagnostic expert.
If you're stuck — router down, breaker tripped, kid-locked door — just go-live.
Repo: https://t.co/kj2KcaynQd
Try it: point your camera, talk naturally, interrupt anytime.
Built with **Gemini Live** + **Google Cloud Run**.
#GeminiLiveAgentChallenge