# Syed Yousuf Faizan — Full Unabridged Documentation & System Logs Archive

> Full Stack & Generative AI Developer specializing in MERN stack, LangChain, LangGraph, Multimodal LLMs, Three.js, Modal.com GPU inference, and Cloudflare R2 serverless architectures.

Syed Yousuf Faizan is a full-stack engineer and AI developer based in Bengaluru, India. He builds production-grade AI applications, video processing pipelines, serverless GPU infrastructure, and high-performance web systems.

## Developer & System Contact
- **Name:** Syed Yousuf Faizan
- **Role:** Full Stack & Generative AI Developer
- **Location:** Bengaluru, Karnataka, India
- **Email:** syedyousuffaizan@gmail.com
- **Website:** https://yousuffaizan.com
- **GitHub:** https://github.com/SyedYousufFaizan
- **LinkedIn:** https://linkedin.com/in/syedyousuffaizan
- **X (Twitter):** https://x.com/Retro_Nerve

---

## 1. Technical Stack & Architecture Overview

### Frontend & Web Visualization
- **Core Framework:** React 19, Next.js (App Router v15/v16 Server Components & Metadata API)
- **Languages:** TypeScript, JavaScript (ESNext)
- **Styling:** Tailwind CSS v4, Vanilla CSS design systems, Glassmorphism, Industrial Brutalist & Retro OS Themes
- **3D & Motion Graphics:** Three.js, React Three Fiber, WebGL shaders, GSAP ScrollTrigger (word-by-word scrubbing & layout pinning)

### Backend & Cloud Infrastructure
- **API Runtimes:** Node.js, Express.js, Python systems programming, C, C++
- **Interface Protocols:** RESTful APIs, GraphQL, Server-Sent Events (SSE), WebAssembly
- **Serverless Compute:** Modal.com (on-demand A100 & RTX PRO 6000 GPU cluster inference)
- **Object Storage & CDN:** Cloudflare R2 (zero egress fee S3 bucket storage & binary data streaming)
- **Deployment Automation:** Vercel (Frontend & Edge Functions), Render (API Microservices)

### AI Engines & Data Stores
- **Agent Orchestration:** LangChain, LangGraph, custom tool-calling agents
- **Multimodal Models:** Qwen2.5-VL-72B (vision-language frame analysis), DeepSeek V3 (reasoning & structured output), Mistral LLM, OpenAI Whisper
- **Data & Vector Stores:** MongoDB NoSQL, Vector Databases (semantic embedding indices & RAG pipelines)
- **Prompt Engineering:** Strict JSON schema output, system prompt grounding, hallucination mitigation

---

## 2. Featured Software Projects

### Project 1: Axiom — AI-Powered Neurological Video Analyzer
- **Description:** Axiom processes video content, runs GPU inference to predict neurological brain activation in near real-time, and visualizes it as an interactive 3D brain synchronized with video playback.
- **Core Problem:** Standard analytics (views, watch time) are downstream proxies. They don't measure real-time neural engagement during consumption.
- **AI Core:** Runs Meta's TRIBE v2 fMRI foundation model on Modal.com GPUs (A100 / RTX Pro 6000), outputting brain activation across 20,484 cortical vertices on the fsaverage5 brain mesh every 1.49 seconds.
- **Attention Signal:** Combines Dorsal Attention Network (DAN) + Ventral Attention Network (VAN) minus Default Mode Network (DMN) suppression.
- **Compression & Streaming:** Brain activation data is converted to float16 binary (50% compression), stored on Cloudflare R2, and streamed to the browser.
- **3D Visualization:** Rendered in Three.js at 60fps with vertex color interpolation on an fsaverage5 brain surface.
- **AI Edit Pipeline:** Extracts video frames via OpenCV, describes them using Qwen2.5-VL-72B, fuses transcript + attention curves, and prompts DeepSeek V3 to output timestamp-specific edit recommendations.

### Project 2: SmartClause — AI-Powered Contract Analyzer
- **Description:** Automates legal contract analysis by parsing PDF, DOCX, and TXT documents, extracting clauses, and classifying risk severity (Low, Medium, High).
- **Core Architecture:** Built on MERN stack (MongoDB, Express, React, Node.js) with Mistral LLM APIs and custom prompt engineering for risk tier classification.

---

## 3. Published Technical Articles & System Logs

### Article 1: Axiom: I Built a Tool That Shows You How the Human Brain Responds to Your Content
*URL:* https://yousuffaizan.com/blog/axiom
*Published Date:* 2026-07-19
*Author:* Syed Yousuf Faizan

#### Full Unabridged Text:

Content creation is still a black box, no matter how many analytics you have. You never truly know what your audience feels when they watch your content. And what kind of content is 'popping' at the given time, is still a mystery.

I wanted to know what was actually happening inside the viewer's brain.

So I built **Axiom**.
It all started when i came across a video on Meta's TribeV2 model (*A Predictive Foundation Model Trained to Understand How the Human Brain Processes Complex Stimuli*) and thats when i exactly knew what i was gonna do for the next few weeks.
**Axiom** is a neurological content analysis platform that uses real fMRI-based AI to predict brain activation across every moment of a short-form video, visualise it in 3D, and tell you exactly what to change and why.

#### The Problem With Content Analytics Today

Every platform gives you the same metrics. Views, watch time, completion rate, saves. These are downstream signals — they tell you *what happened* after the content has been uploaded and consumed, and not really what happens in between, whether it will be a hit or miss.

A video can have a 70% completion rate and still be losing the viewer's brain 15 seconds in. The algorithm rewards the output. Nobody measures the input — which is the brain itself.

That's what Axiom tries to change.

#### What Is TRIBE v2?

Axiom is powered by **Meta's TRIBE v2** — a multimodal AI model trained on real fMRI brain scan data. TRIBE v2 was built by Meta's research team to predict neurological responses to video content.

Here's what makes it unusual: it wasn't trained on engagement proxies. It was trained on actual brain scans where real people watched videos inside an MRI machine and researchers measured blood oxygenation changes across the cortex. TRIBE v2 learned to predict those responses from the video content alone, and the whole model itself is built upon 3 pretrained AI models — one for visual, one for audio, and one for transcription/text — fused via a transformer architecture.

The model outputs activation predictions across **20,484 vertices on the fsaverage5 brain surface** — a standard neuroimaging mesh used in academic research. One prediction per vertex, per 1.49 seconds of video. That's the hemodynamic resolution of fMRI.

The result is a dense, spatially structured map of predicted brain engagement for every moment of your content.

#### What Axiom Does (When you Upload a Reel)

1. **Neurological Inference:**
   The video is sent to an A100 or an RTX PRO 6000 GPU hosted on Modal.com where the TRIBE v2 model runs inference. The model processes visual frames, audio, and speech simultaneously outputting a brain activation map across all 20,484 cortical vertices for every 1.49-second window of the video. This takes roughly 3–5 minutes for a 60-second Reel.

2. **Attention Timeseries:**
   Axiom extracts an attention signal by combining activation from the **Dorsal Attention Network** (voluntary focused attention) and **Ventral Attention Network** (stimulus-driven attention capture) against suppression of the **Default Mode Network** (mind wandering).
   Equation: `(DAN + VAN) - DMN = Attention Score`
   The result is a second-by-second attention curve for the entire video with Hook score, Peak moment, Drop point, and High attention percentage metrics.

3. **Live 3D Brain Visualisation:**
   The brain activation data is serialised as a float16 binary array (converted from float32 format), resulting in 50% compression with negligible data loss. This data is served from Cloudflare R2. Three.js renders an fsaverage5 inflated brain surface mesh with 40,004 vertices, updating vertex colors at 60fps in sync with video playback.

4. **AI-Powered Edit Suggestions:**
   Axiom extracts video frames using OpenCV, describes them using **Qwen2.5-VL-72B**, transcribes speech with Whisper, fuses attention scores + visuals + transcript into a single timeline, and passes it to **DeepSeek V3** for structured, timestamped edit recommendations.

#### System Architecture & Stack

```
Browser
  ↓ multipart stream (busboy)
Node.js backend (Render)
  ↓ stream directly to
Cloudflare R2
  ↓ presigned video URL
Modal A100/RTX PRO 6000 GPU (on demand)
  ↓ TRIBE v2 inference
  ↓ brain.bin + report.json streamed back to
Cloudflare R2
  ↓ presigned URLs returned to
Browser
  ↓ fetches directly from R2 CDN
```

- **AI Inference:** Meta TRIBE v2
- **GPU Compute:** Modal.com (A100, RTX PRO 6000)
- **Storage & CDN:** Cloudflare R2
- **Vision Analysis:** Qwen2.5-VL-72B (OpenRouter)
- **Suggestions:** DeepSeek V3
- **Transcript:** OpenAI Whisper
- **3D Rendering:** Three.js, React Three Fiber
- **Frontend:** Next.js, Tailwind CSS
- **Backend:** Node.js, Express
- **Cost per Reel:** ~$0.56 total compute cost