Skip to content
View PremPastagia's full-sized avatar
1 follower · 1 following
  • India

Block or report PremPastagia

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
PremPastagia/README.md

Hi, I'm Prem Pastagia 👋

Systems & Machine Learning Engineer | Distributed Systems (C++20), Production MLOps & Applied AI

┌────────────────────────────────────────────────────────┐
│  Distributed Systems • High-Throughput MLOps • RecSys  │
└────────────────────────────────────────────────────────┘

🚀 Featured Engineering Projects

  • ⚙️ Distributed Job Engine — C++20 distributed job processing and workflow engine featuring thread-safe priority scheduling, exponential-backoff retries, per-job deadlines, DAG workflows, SQLite durability with group commit, and lease-based multi-process execution with fencing tokens.
  • 📈 ML Serving & Monitoring Platform — End-to-end MLOps platform: automated data validation, reproducible training pipelines, MLflow model registry, FastAPI serving (256 req/s, 4.7ms p95), Prometheus metrics, statistical drift detection, and gated retraining with verified rollback.
  • 🧠 Grounded RAG Knowledge System — High-precision retrieval-augmented generation over IETF RFCs with hybrid BM25 + dense vector retrieval and verified citation grounding.
  • 🎯 RecSys From Scratch — Five recommendation paradigms (UBCF, IBCF, regression-based neighbourhood CF, SLIM, graph propagation) built from mathematical first principles on MovieLens 100K without recommender libraries.
  • 🖐️ Gesture Control — Low-latency Windows virtual desktop manager driven by real-time webcam hand gestures, MediaPipe computer vision, and native OS shortcuts.
  • 🔺 TriForm Tactics — 4-player nested strategy board game powered by in-browser neural AI agent (onnxruntime-web), React 19, and Vite.
  • 🦐 SensoPack — IoT & AI smart packaging system combining colorimetric bioactive film, gas/temp sensor telemetry, and machine learning for perishable food freshness prediction.

🛠️ Technical Toolkit

  • Languages: C++20, Python 3.12, TypeScript, JavaScript, SQL
  • Systems & Backend: POSIX Concurrency, Multithreading, SQLite, FastAPI, Node.js, Express, Docker
  • ML & Data: PyTorch, scikit-learn, XGBoost, LightGBM, MLflow, Prometheus, NumPy, SciPy, ONNX Runtime
  • Frontend & Mobile: React 19, Next.js, React Native / Expo, Vite, Tailwind CSS

📬 Connect With Me

Pinned Loading

  1. Auto_Caller_KTJ Auto_Caller_KTJ Public

    Turborepo monorepo for festival tele-calling operations, automated outreach orchestration, and fraud prevention with Expo mobile app and Next.js dashboard.

    TypeScript

  2. distributed-job-engine distributed-job-engine Public

    C++20 distributed job processing and workflow engine: thread-safe priority scheduling, exponential-backoff retries, per-job deadlines, DAG workflows, SQLite durability with group commit, and lease-…

    C++

  3. Gesture-Control Gesture-Control Public

    Low-latency Windows virtual desktop switcher and window controller driven by webcam hand gestures, MediaPipe computer vision, and native OS shortcuts.

    Python

  4. ml-serving-monitoring-platform ml-serving-monitoring-platform Public

    End-to-end MLOps platform: data validation, reproducible training, MLflow registry, FastAPI serving (256 req/s, 4.7ms p95), Prometheus monitoring, statistical drift detection, gated retraining with…

    Python

  5. rag-knowledge-system rag-knowledge-system Public

    Grounded RAG over IETF RFCs: hybrid retrieval, verified citations, and a measured decision to disable the reranker. Every CV number resolves to a results file.

    Python

  6. recsys-from-scratch recsys-from-scratch Public

    Five recommendation paradigms (UBCF, IBCF, regression-based neighbourhood CF, SLIM, graph propagation) implemented from scratch on MovieLens 100K — no recommender libraries. Benchmarked on accuracy…

    Python