- 🎓 CS student at Penn State Harrisburg
- 💼 Former SWE Intern @ Clark Associates → built CI/CD pipelines, Kubernetes migrations, GitOps automation
- 🛠️ Passionate about full-stack apps, scalable DevOps systems, and AI-powered products
- 🌱 Currently learning AI, advanced system design, and scaling microservices
- ⚡ Motto: “Building today what others will use tomorrow.”
Languages: C++, Python, TypeScript, SQL, Java
Frameworks/Libraries: React, Node.js, Express, Next.js, LangChain
Cloud & DevOps: AWS, Docker, Kubernetes, ArgoCD, Terraform, Azure DevOps, CI/CD pipelines
Databases: PostgreSQL, MongoDB, MySQL, Redis
Other Tools: Git, Postman, SonarQube, Firebase, Jenkins, Linux
🔹 Meter – Usage-based monetization gateway for MCP servers
- Transforms any MCP server into a paid endpoint with per-tool pricing enforced via MNEE payments
- Agents submit payment proof before execution no custody, no billing stack, direct wallet settlement
- Built with Node.js/Express backend, Next.js dashboard, Prisma ORM, and LangChain MCP integration
- Includes analytics dashboard showing per-tool usage, revenue trends, success rates, and live on-chain wallet balance
- Enables agent-to-agent commerce and autonomous service monetization without custom billing infrastructure
🔹 AI Photo Generator – Train AI models on your own photos
- Upload 10–20 personal photos to train custom LoRA models (via Fal.ai)
- Generate studio-quality portraits from text prompts
- Built with Next.js + Express (Bun) in a Turborepo
- Features webhook updates, R2 cloud storage, and a personalized dashboard
🔹 DreamPlay – AI powered kids game generator
- Turns simple ideas into fully playable HTML5 games using custom engines for collect, falling, and jump mechanics
- Generates icons, lessons, and dynamic game configs with Claude Sonnet and a rule based refinement pipeline
- Built with Next.js, Turborepo, Prisma, and a full agent workflow that filters vague or unsafe prompts
🔹 FWD Sedan Diagnostic Triage - AI-powered auto repair diagnostics
- Fine-tuned LoRA adapter (355M trainable parameters on ERNIE-4.5-21B) trained on 1,000 ChatML samples for shop-style triage of FWD sedan idle roughness, vibration, and AC-load symptoms
- Delivers concise 2-5 sentence diagnostics with mileage-weighted cause ranking instead of generic checklists, optimized for busy mechanics and car owners
- Trained end-to-end on single A100 GPU in 10 hours using Unsloth + TRL SFTTrainer with 4-bit quantization
🔹 Perkora – AI-powered perks discovery for startups
- Sign up with email or OAuth and answer intelligent questions about your startup
- Enter your stage, industry, team size, and burn rate to unlock personalized perk matches
- Raindrop AI agent searches across 100+ perk databases and ranks results with Smart Inference
- SmartMemory maintains conversation context while SmartSQL stores your profile and discovered perks
- Built with Next.js frontend powered by Raindrop, SmartSQL, SmartMemory, and Smart Inference engine
- Get curated recommendations with direct apply links, approval tracking, and saved favorites dashboard
🔹 Micrograd Engine – Autodiff engine built from scratch
- Implemented a full reverse mode autodiff system with a Value class, computation graph tracking, and custom backward functions
- Built Neuron, Layer, and MLP modules from first principles to train simple neural networks without any deep learning frameworks
- Added visualization support for computation graphs and a PyTorch comparison script to validate gradient correctness
- 🌐 Portfolio
- ✉️ Email: mxs7142@psu.edu
