╔═════════════════════════════════════════════════════╗
║ ║
║ █████╗ ██████╗ ██╗ ██╗ █████╗ ███╗ ██╗ ║
║ ██╔══██╗██╔══██╗╚██╗ ██╔╝██╔══██╗████╗ ██║ ║
║ ███████║██████╔╝ ╚████╔╝ ███████║██╔██╗ ██║ ║
║ ██╔══██║██╔══██╗ ╚██╔╝ ██╔══██║██║╚██╗██║ ║
║ ██║ ██║██║ ██║ ██║ ██║ ██║██║ ╚████║ ║
║ ╚═╝ ╚═╝╚═╝ ╚═╝ ╚═╝ ╚═╝ ╚═╝╚═╝ ╚═══╝ ║
║ ║
║ Building AI that works for everyone. ║
╚═════════════════════════════════════════════════════╝
Birla Global University, Bhubaneswar · Batch 2025–27 · GPA: 8.18
"The development of the nation and betterment of society."
I started coding seriously in early 2026. In ~2 months: 13+ repos, 5 live projects, 1 production Android app, and an active research problem on on-device AI. I build backend-first, plan architecture in a physical notebook during weekdays, and execute on weekends.
My work focuses on:
- Edge AI & on-device inference
- AI-powered security tooling
- Backend-first application architecture
- Civic-tech platforms
- AI systems for accessibility
- Production-focused full-stack engineering
+------------------------------------+
| PAT AI |
| IND * Indian AI Innovation Group |
| (Long-term Vision) |
+-------+----------+-----------------+
| |
+------------+ +----------------+
| |
v v
+------------------+ +------------------+
| PAT_7 | | AI Defender |
| Multi-AI Core | | Security Plat. |
| Build Platform | | [In Progress] |
+------------------+ +------------------+
|
v
+------------------+
| Darshan-3 |
| Edge Vision AI |
| [Active] |
+------------------+
Standalone Projects:
+------------------------------------------+
| PS Predictor * PTJP (Android) |
| CivicConnect (Vercel) |
| Image Caption Generator (GitHub) |
+------------------------------------------+
An Indian AI Innovation Group building a multi-model AI platform focused on integrating AI-assisted workflows, project systems, and scalable backend architecture.
Core Components:
-------------------------------------------------------------
PAT_7 --> Multi-AI core brain & build platform
AI Defender --> Security analysis platform
Darshan-3 --> Edge vision AI for accessibility
-------------------------------------------------------------
* Multi-AI interaction workflows
* Modular backend structure
* Scalable platform-oriented architecture
* Production deployment on Vercel
-------------------------------------------------------------
Stack: Python TypeScript Next.js Vercel
Targeting individual developers & small teams
Module Pipeline:
-------------------------------------------------------------
[M1: Data Gen] --> [M2: Static Scanner] --> [M3: Runtime]
DONE IN PROGRESS 43 JSONL files
Groq + Llama-3.3 Flask / Express.js MITRE ATT&CK
70b-versatile OWASP Top 10 8 Attack Categories
-------------------------------------------------------------
|
+---------+----------+
v v
VS Code Extension CI/CD Integration
(Distribution) GitHub Actions / Jenkins
Shared REST API
Stack: Python Groq API Llama-3.3-70b Flask OWASP MITRE ATT&CK
On-device. Offline. Sub-100ms. No cloud.
Hardware: Raspberry Pi + Wearable Camera
|
v
+-------------------------------------+
| YOLOv8 Nano --> Object Detection |
| OpenCV --> Frame Processing |
| pyttsx3/gTTS --> Audio Guidance |
+-------------------------------------+
|
Hard Requirements:
* < 100ms inference latency
* Zero cloud dependency
* Fully on-device (edge-only)
YOLOv8 Nano OpenCV Python Raspberry Pi pyttsx3
System Processes --> BufferRegistry (per-PID rolling windows)
|
v
+------------------------------+
| Hybrid Decision Engine |
| RF Ensemble + SVM Model |
| + Rule-Based Override |
+------------------------------+
|
Flask Web Dashboard
setup.bat / start.bat
scikit-learn Random Forest SVM Flask Python
Multi-role Auth System:
+----------+ +-----------+ +-------------+ +------------+
| Citizen | | Volunteer | | Institution | | Government |
+----+-----+ +-----+-----+ +------+------+ +-----+------+
+---------------+----------------+--------------+
|
Supabase (PostgreSQL + RLS)
|
Next.js PWA --> Vercel [LIVE]
Next.js Supabase PostgreSQL RLS TypeScript
Job Lifecycle State Machine:
+------+ +---------+ +-----------+
| OPEN | --> | ONGOING | --> | COMPLETED |
+------+ +---------+ +-----------+
Stack: React Native (Expo) + TypeScript + Supabase + EAS Build
Target: Production Android [DEPLOYED]
React Native Expo TypeScript Supabase
Image Input --> ResNet-50 (Encoder) --> LSTM (Decoder) --> Caption
|
HuggingFace Transformers (fallback)
Trained locally on COCO dataset
Flask API . GitHub deployed
PyTorch ResNet-50 LSTM HuggingFace Flask COCO
+-------------------------------------------------------------+
| Active Research Problem |
| |
| Problem: LLM-dependent apps fail on low-end devices |
| |
| Approach: |
| * Domain-specific knowledge distillation |
| * Security-focused RAG pipelines |
| * ~3B parameter models matching larger model performance |
| |
| Motivation: Darshan-3 requires on-device inference |
| with sub-100ms latency on Raspberry Pi hardware |
| |
| Impact: Makes AI accessible on constrained hardware |
| because not everyone has a GPU. |
+-------------------------------------------------------------+
- 🔨 Building AI Defender Module 2 (static scanner) & Module 3 (runtime monitor)
- 🔬 Researching on-device LLM distillation for Darshan-3 edge deployment
- 🚀 Expanding PAT AI ecosystem — PAT_7 core platform
- 💼 Interning at SPK-MAK Technologies as Web Developer
- 📖 MCA Year 1 @ Birla Global University · GPA: 8.18
+-----------------------------------------------------+
| Open to: Collaborations . Research Partnerships |
| Internship Opportunities . Open Source |
+-----------------------------------------------------+
"Not everyone has a GPU. Building AI that doesn't care."