Search for a command to run...
No comments yet. Be the first to comment.
Privacy isn't dead. It's just being rebuilt. The old internet promised us "move fast and break things." We moved fast. We broke things. Now we're realizing what we broke: privacy. Every LLM you use wa
Coding is changing faster than you think. Not just "AI helps with autocomplete." AI is becoming the primary developer. You become the reviewer. The Skill Inversion Old model (2020): Human writes code.
Single AI agents are limited. They make mistakes. They miss context. They take wrong paths. Multi-agent systems don't have these problems. They have different problems. But better ones. The Single Age
You've learned basic RAG. You embed documents. You retrieve them. You feed them to an LLM. It generates an answer. Then you deploy it. And it breaks. Because the real world isn't simple retrieval. Adv
The biggest AI models aren't winning anymore. GPT-4: 1.8 trillion parameters Claude 3 Opus: 200+ billion parameters Claude 3.5 Sonnet: Unknown but smaller Llama 3.1: 405 billion parameters PhiMoe (ope
Muhammad Zulqarnain | Full Stack AI Engineer & Geospatial Developer
16 posts
A blog by Muhammad Zulqarnain — Full Stack AI Engineer & Geospatial Developer based in Turku, Finland. I write about RAG systems, LLMs, Prompt Engineering, Next.js, TypeScript, and geospatial development. Practical insights, deep dives, and real-world AI solutions.
The era of "text-in, text-out" AI is ending.
Multimodal models—AI that understands images, video, audio, and text together—aren't the future. They're the present. And they're about to transform entire industries.
Two years ago:
ChatGPT: text only
DALL-E: text to image
Whisper: audio to text
Separate tools for separate tasks
Today:
Claude 3.5 Sonnet: reads images, documents, charts
GPT-4o: understands video, images, audio
Gemini 2.0: 1-hour video understanding
Everything talks to everything
Single unified models that understand the world like humans do.
1. It Closes the AI Perception Gap
AI could read text. But it was blind.
A company had a problem: their customer support team spent 2 hours per ticket understanding issues. Why? Because customers sent photos, videos, and messages.
Textual AI couldn't help. You'd feed it: "My door won't open. [image]"
AI reads text: "Hmm, needs more info."
Multimodal AI reads image + text: "That's a broken hinge. Here's the fix."
Same ticket: 5 minutes. Solved.
2. Documents Are More Than Text
Your company has:
Contracts with signatures
Invoices with tables
Flowcharts
Diagrams
Screenshots
Handwritten notes
Text-only LLMs struggle. They hallucinate. Misread tables. Fail on diagrams.
Multimodal AI reads the actual document. Layout, images, tables, all included.
Accuracy jumps from 70% to 95%.
3. Video is the New Data Frontier
Fact: There's more video footage recorded every 48 hours than existed in 1990.
Nobody is watching all of it. Understanding it requires AI.
Multimodal AI can now:
Watch a manufacturing video and spot quality issues
Review security footage and flag suspicious activity
Analyze medical imaging videos and assist diagnosis
Understand educational content and summarize it
Security firms are already using this. Watching 100 camera feeds with AI instead of 5 humans.
Retail:
Computer vision identifies shelf gaps in real-time
AI calls stock adjustments to the warehouse
Cost: $2K for camera setup
Saves: $50K/year in inventory optimization
Healthcare:
Multimodal AI reads X-rays + patient history
Suggests diagnoses faster than radiologists alone
Radiologists use AI as a second opinion
Catch 15-20% more early-stage issues
Manufacturing:
Factory camera + AI watches assembly lines
Spots defects before they ship
Savings: 5-15% reduction in returns
Insurance:
Claimant submits: description + photos of car damage
AI estimates damage cost in 2 minutes
Used to take 3 days + adjuster visit
Some engineers say: "LLMs just got good. Why rebuild everything?"
Because:
Text can't capture reality. Someone describes: "My knee hurts when I walk." Multimodal: Watches them walk. Sees the limp. Biomechanics shows it's knee, not hip. Textual: "Could be many things."
Text loses information. Document: "Revenue grew this quarter." Chart in that document: Revenue actually declined 30%, but one product grew 200%. Text LLM misses it. Multimodal AI sees both.
Text alone is slow for complex work. A lawyer reading a 500-page contract: 8 hours. Multimodal AI: 30 seconds. Points to risky clauses. Suggests alternatives.
1. Real-Time Video Understanding AI watching live camera feeds. Not batch processing recorded video. Live.
Your storefront: live footfall, customer flow analysis
Factory: real-time defect detection
Hospitals: patient monitoring with real-time alerts
2. 3D Scene Understanding AI that understands spatial relationships:
"Show me all chairs in this image"
"What's the angle of that camera?"
"How much space is between these objects?"
Not just "I see objects." But "I understand the 3D scene."
3. Audio + Vision Integration You won't just transcribe calls. AI will:
Watch the video of a sales call
Hear the tone
Read facial expressions
Understand context from body language
Better coaching for sales teams
4. Multimodal Reasoning AI that reasons across modalities: "Based on the image of the equation AND my knowledge of physics, here's the solution." Not separate models. Not chained tasks. Unified reasoning.
Every company's data is trapped in silos:
Text in Slack, emails, documents
Images in folders
Video in archives
Audio in call recordings
Companies that:
Organize their multimodal data
Train or fine-tune on it
Build internal multimodal tools
...Will have superhuman advantages over competitors.
Your rivals are still searching through documents manually. You: "AI, summarize all issues from customer videos this month."
Answer: 30 seconds.
Competitive moat: Huge.
If you're building:
Audit your data. How much is text? Images? Video?
Identify one task that's multimodal (text + image, usually)
Try Claude 3.5 Vision or GPT-4o Vision on it
Build a prototype with those tools
You'll be shocked at what's possible
If you're hiring: Look for engineers who understand:
Vision transformers
Prompt engineering for vision
How to structure multimodal data pipelines
They're rare. But they're worth 3x a regular LLM engineer.
If you're investing: Multimodal infrastructure companies are coming:
Tools for organizing visual data
Platforms for training multimodal models
APIs that make multimodal workflows easy
Earlier stage than text LLMs. More upside.
Text AI proved that language models could reason.
Multimodal AI proves that AI can perceive.
Combine reasoning + perception, and you get something close to intelligence.
Not human intelligence. But useful intelligence.
The 2020s will be about companies that figure out how to give their AI eyes and ears.
The rest will compete on text alone.
Guess who wins?ing the World
