|
YOLO · ONNX · PyTorch Model optimization, browser-side inference and computer vision experiments. |
Vitis AI · DPU · FPGA Deployment-focused work on ZCU104, Kria KR-260 and quantized inference. |
Linux · Docker · CI/CD Reproducible services, deployment pipelines and homelab-style systems. |
📡 Click to expand my GitHub signal map
GitHub focus:
> AI product engineering
> Edge AI deployment experiments
> DevOps utilities and system fixes
> Full-stack event and API platforms
I’m Partha, an AI/ML and DevOps-focused IT engineering student building practical systems across:
- Edge AI and FPGA-accelerated inference
- YOLO model deployment on DPU boards
- Browser-side AI with ONNX Runtime Web
- Linux, Docker, GitLab CI/CD and homelab infrastructure
- Full-stack applications with React, Next.js, Node.js and SQL
- Cloud AI workflows using AWS Bedrock, Polly, DynamoDB and Amplify
I enjoy working at the intersection of hardware-aware AI, deployment engineering, and user-facing AI products.
🧠 Click the emoji to decode my dev mode
I like projects where:
> AI has to run under real constraints
> deployment matters as much as the model
> users get a useful workflow, not just a demo
name: Partha / PG username: Partha-dev01 role: AI/ML & DevOps Engineer education: B.Tech in Information Technology primary_domains: - Edge AI - Computer Vision - DevOps - Full-Stack Engineering - Cloud AI - Linux Systems current_focus: - YOLO inference optimization - Vitis AI and DPU deployment - Browser-side ONNX inference - Local LLM and AIOps experiments - Dockerized deployment pipelines engineering_style: - Build practical systems - Keep deployments reproducible - Optimize for real-world constraints - Learn close to hardware and users
|
I work with vision models, local inference, ONNX deployment, quantization and model optimization.
|
I’m interested in making models run efficiently on constrained hardware and FPGA/DPU platforms.
|
|
I like building reliable environments where software can be tested, deployed and reproduced.
|
I build web platforms, dashboards, APIs and cloud-backed AI applications.
|
|
AI-enabled early autism screening, diagnosis support and post-diagnosis care platform. Built around privacy-first browser-side AI workflows, multimodal scoring, clinical report generation and cloud synchronization. Stack: Next.js, TypeScript, ONNX Runtime Web, AWS Bedrock, Polly, DynamoDB, Amplify Highlights:
|
FPGA/DPU-focused deployment work for optimized pedestrian detection. Refactored YOLOv11-Nano to remove unsupported DPU operators, then quantized, compiled and deployed the model using Vitis AI. Stack: Vitis AI, PyTorch, ZCU104, PYNQ, DPU, Quantization Highlights:
|
|
Local agent-based AIOps system for telemetry annotation and root-cause analysis. Designed dual local Qwen agents with constitutional safety checks, graph memory and human approval workflows. Stack: Python, FastAPI, React, Neo4j, Docker Highlights:
|
Mobile app concept for farmer-wholesaler market interaction. Focused on live bidding, market analytics and better price discussion workflows between farmers and wholesalers. Stack: React Native, Firebase Highlights:
|
|
Convert a real problem into a focused technical objective. |
Build the fastest working version before over-engineering. |
Improve speed, memory, model size and deployment compatibility. |
Package for browser, cloud, container or edge board. |
Test under real constraints and improve the loop. |
🧠 AI / ML Deployment Track
Input Data
↓
Model Training / Fine-tuning
↓
ONNX / PyTorch Export
↓
Quantization + Compatibility Fixes
↓
Runtime Target: Browser / Cloud / DPU / Edge Board
↓
Benchmarking + Validation
⚙️ DevOps / Systems Track
Local Development
↓
Git + Version Control
↓
Dockerized Environment
↓
CI/CD Pipeline
↓
Cloud / Homelab Deployment
↓
Logs, Monitoring and Feedback
🌐 Full-Stack Product Track
User Flow
↓
Frontend UI
↓
Backend API
↓
Database / Cloud Services
↓
Authentication + Role-based Access
↓
Deployment + User Testing
Build fast. Optimize deeply. Deploy cleanly. Validate honestly. Repeat.
# Things I enjoy building and maintaining linux_homelab="Proxmox + containers + reproducible services" ci_cd="GitLab pipelines, Dockerized workflows, deployment automation" systems="RHEL, Bash, networking basics, local environments" cloud_ai="AWS Bedrock, Polly, DynamoDB, Amplify" engineering_goal="Make projects easy to run, test, deploy and improve"
Red Hat Certified System Administrator
Platform: RHEL 9.3
Focus: Linux administration, users, permissions, services, storage and system operations
|
Vitis AI → DPU → YOLO Optimizing computer vision models for FPGA/DPU inference. |
ONNX Runtime Web Running privacy-first AI inference directly inside the browser. |
Docker → CI/CD → Deploy Packaging projects into reproducible deployment workflows. |
Logs → Agents → RCA Exploring local LLM agents for telemetry and root-cause analysis. |
🧩 Open detailed quest log
Edge AI Quest
> Refactor YOLO models for DPU compatibility
> Quantize, compile and validate on FPGA boards
> Benchmark performance under real constraints
Browser ML Quest
> Keep inference local and privacy-first
> Run ONNX models with Web Workers
> Build usable AI flows without server-side model dependency
DevOps Quest
> Containerize applications cleanly
> Automate repeatable build and deployment paths
> Keep projects easy to run, test and improve
AIOps Quest
> Use local agents for telemetry annotation
> Connect logs, traces and graph memory
> Support human-reviewed root-cause analysis
> How do we make AI fast enough for edge hardware?
> How do we keep AI apps private and browser-first?
> How do we deploy full-stack systems without fragile manual steps?
> How do we combine local LLMs with real operational workflows?
> How do we turn research prototypes into usable systems?
while learning:
build something practical
test it under constraints
document what worked
improve the deployment path
repeat with a harder problem
Building AI that runs closer to the user, closer to the hardware, and closer to real-world impact.