Skip to content
View Uditp11's full-sized avatar

Block or report Uditp11

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. 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
Uditp11/README.md
Udit Parihar. I build machine learning models I can measure, and the APIs, web apps and tests around them. Frankfurt am Main. M.Sc. Global Software Development, Hochschule Fulda.

Open to Werkstudent, internship and part-time roles in Germany: ML and AI engineering, Java and Python backend, full-stack, QA and test automation, data and automation.

Projects, one measured result each

cifar10-from-scratch: 96.06% test top-1, evaluated once. ResNet-18 written from scratch in PyTorch. 10,000 held-out images. Every recipe choice tested in a logged ablation. applytrack: 86 tests, 0 skipped in CI. Java and Spring Boot API. A status state machine refuses illegal moves. Kubernetes deploy in CI, Terraform deploy to Azure. vectordb-from-scratch: 3 search algorithms, by hand. HNSW, KD-Tree and brute force in Python and NumPy. Tests prove KD-Tree matches brute force exactly. Recall and latency benchmarked, local RAG. solar-inspect, v1.0: 5 modules, each test split read once. Solar panel inspection in PyTorch: U-Net from scratch, YOLOv8 detection, Kalman and SORT, defect classifier, plant analytics with SQL. shift-log, live: end-to-end tested in 2 browsers. Offline-first phone app my friends and I use. Playwright in WebKit and Chromium, one branch per task, tests run before every deploy. Shree_Shyam_Constructions, live: website for a paying client, in English and Hindi. Next.js, TypeScript, Tailwind CSS.

Tests run on every push:

applytrack cifar10 vectordb shift-log einkaufzentrum
applytrack CI cifar10 CI vectordb tests shift-log test and deploy einkaufzentrum tests

Pick your lane

ML and AI engineering
  • solar-inspect: five-stage solar panel inspection in PyTorch. U-Net segmentation written from scratch, YOLOv8 detection with IoU, NMS and AP written by hand, Kalman and SORT tracking checked on synthetic flyovers, defect classification with calibration and distillation, and analytics on real plant data. Each of the three image test splits read once. Open the headline results
  • cifar10-from-scratch: ResNet-18 written from scratch, 96.06% test accuracy. Open the ablation ledger
  • vectordb-from-scratch: HNSW and KD-Tree written by hand, local RAG with Ollama. Open the benchmark
  • RAG_Adaptive (team project): LangGraph chatbot that routes each question to documents, the LLM or web search, then grades and retries retrieval
Java and Python backend
  • applytrack: Java and Spring Boot API, PostgreSQL with Flyway migrations, a status state machine that refuses illegal moves, statistics computed in SQL
  • vectordb-from-scratch: async FastAPI layer over a hand-written vector index
  • einkaufzentrum (team course project): Flask and SQLAlchemy API with JWT roles and integration tests in CI
Full-stack
QA and test automation
  • shift-log: Playwright end-to-end tests in WebKit at iPhone size and Chromium at Android size, run before every deploy. Open playwright.config.js
  • applytrack: JUnit 5, Mockito and Testcontainers against a real PostgreSQL
  • cifar10-from-scratch: pytest on a Python 3.10 and 3.12 CI matrix
  • solar-inspect: 232 pytest tests, checked against hand-computed values and against torchvision and statsmodels
  • einkaufzentrum (team course project): three integration suites, run in CI
Data and automation
  • applytrack: response rate and status counts computed in SQL
  • solar-inspect: regression on real solar plant data with Newey-West errors, and five SQLite queries with window functions, each checked against pandas. Open sql/
  • einkaufzentrum (team course project): SQL layer through SQLAlchemy, SQLite and MySQL
  • Automation (n8n, Power Automate, Python and JavaScript scripts) was used at work
  • Private, not public: my own job-hunt agents, built with Claude, collect postings every day, score them and file them in Notion. I also wrote Claude Code skills and my own MCP server

Stack

Every badge below is used in a public repo on this profile.

ML and AI   Python PyTorch NumPy pandas LangChain

Backend and data   Java Spring Boot FastAPI Flask Node.js PostgreSQL SQL SQLite

Frontend   JavaScript TypeScript React Next.js

Testing   pytest JUnit 5 Playwright

Cloud and delivery   Docker Kubernetes Terraform GitHub Actions

How I work

Habit Where to see it
Measure before I claim cifar10 ablation ledger: every recipe choice tested against a measured noise floor
Touch the test set once solar-inspect eval script refuses to run until the config is committed
Write decisions down Decision records in cifar10 (retractions kept) and solar-inspect
One branch per task, tests before deploy shift-log workflow: deploy waits for the tests

Where my skills come from

Area Built in public (open the repos above) Used at work (Kidz Brain Academy, 2023 to 2024) Coursework (graded, M.Sc. and B.Tech)
ML and AI PyTorch, NumPy, computer vision (segmentation, detection with YOLOv8, tracking with Kalman and SORT, classification), knowledge distillation, statistics (regression, Newey-West errors, bootstrap intervals), model evaluation (ablations, calibration, leakage checks), vector search, LangGraph, LangChain, OpenAI API, Ollama OpenAI API in internal tools, with output checks Classical ML and statistics with scikit-learn, PySpark, Hugging Face
Backend and data Java, Spring Boot, PostgreSQL with Flyway migrations, SQLite with window functions, pandas, FastAPI, Flask, SQLAlchemy, REST APIs, JWT Node.js and Python services, REST APIs, SQL, data cleaning Thymeleaf, NestJS
Frontend React, Next.js, TypeScript, Tailwind CSS, offline web apps (service worker, IndexedDB) React Vue 3, Angular
Testing pytest, JUnit 5, Mockito, Testcontainers, Playwright Test cases, manual testing, Postman Two graded testing courses
Cloud and delivery Docker, Docker Compose, Kubernetes (kind), Terraform on Azure, GitHub Actions, Prometheus and Grafana Jenkins
Automation Nothing public yet n8n, Power Automate, Python and JavaScript scripts

Background

  • M.Sc. Global Software Development, Hochschule Fulda (ongoing)
  • B.Tech Computer Science, CGPA 9.06 / 10 · GATE qualified 2022 and 2023
  • Full Stack Developer Intern at Kidz Brain Academy, an education business (Feb 2023 to Aug 2024)
  • English fluent, Hindi native, German A2/B1 (actively improving, uncertified)
Show as a timeline
timeline
    2019 to 2023 : B.Tech Computer Science, CGPA 9.06 / 10
                 : GATE qualified 2022 and 2023
    2023 to 2024 : Full Stack Developer Intern, Kidz Brain Academy, an education business
    2024 onward  : M.Sc. Global Software Development, Hochschule Fulda
    2025         : Team projects, einkaufzentrum and RAG_Adaptive
    2026         : ApplyTrack, cifar10, vectordb, shift-log, solar-inspect
Loading

Contact

Email · LinkedIn

A snake eating my GitHub contribution graph, redrawn every week A small planet with a football orbiting it

Outside code: football and astrophysics.

Pinned Loading

  1. cifar10-from-scratch cifar10-from-scratch Public

    ResNet-18 written from scratch in PyTorch: 96.06% CIFAR-10 test accuracy, each recipe choice backed by an ablation.

    Python 3

  2. solar-inspect solar-inspect Public

    Solar PV inspection pipeline in PyTorch: U-Net segmentation, YOLOv8 detection, Kalman and SORT tracking, defect classification, and plant analytics with triage and SQL. Each test split read once.

    Python 2 1

  3. applytrack applytrack Public

    Job application tracker API in Java and Spring Boot with PostgreSQL, Docker, Kubernetes and a Terraform deploy to Azure

    Java

  4. vectordb-from-scratch vectordb-from-scratch Public

    Vector search from scratch in Python: HNSW, KD-Tree and brute force, with tests, a recall benchmark and local RAG.

    Python 1

  5. RAG_Adaptive RAG_Adaptive Public

    Agentic RAG chatbot in LangGraph: routes each question to documents, the LLM or web search, then grades and retries.

    Python

  6. musterbank-rpa musterbank-rpa Public

    Lernprojekt: fiktives Bank-Backoffice (Flask) und UiPath-Bot fuer Freistellungsauftraege. Nur synthetische Daten.

    Python