Rohit Kumar

AI/ML • Middle • Partially remote, Fully remote • Work permits: India • Citizenship: India • English: B2

Artificial Intelligence Specialist

Python

C++

SQL

JavaScript

Node JS

Active 3 days ago

Invite to interview

Download CVCV

Overview

Technical skills

Projects

Timeline

Roles

Overview

Real-time pose-detection and applied ML developer (Middle level) focusing on Streamlit-based computer vision apps with modular landmark-based detectors. The strongest proven skill is integrating a MediaPipe pose landmarker into a real-time processing loop and building custom landmark heuristics for exercise metrics, as shown in services/vision/exercise_video_processor.py and detectors/biceps_curl.py. There is limited evidence of custom model training pipelines, experiment tracking, production MLOps, or automated test coverage in public code.

Phone

Technical skills

Languages

Python

C++

SQL

JavaScript

Node JS

AI/ML

LLM

NumPy

Pandas

Scikit-learn

OpenCV

Streamlit

Claude

LangChain

LangGraph

PyTorch

NLTK

TensorFlow

MediaPipe

spaCy

Keras

Pillow

Cursor

Groq

AI Agents

OpenAI Codex

AI/ML

Machine Learning

NLP

RAG

Databases

PostgreSQL

SQLite

Supabase

Frontend

Bootstrap

React.js

Frontend

Graphics

DevOps

GitHub

Git

WebRTC

Other

Matplotlib

Express

Seaborn

TensorFlow C++

Text-to-Speech

Computer Vision

CNN

Projects

Digital Signature App

Developed a browser-based digital signature tool with an interactive HTML5 canvas. Added features to draw signatures with adjustable pen and background settings, then clear, save, download, and restore the signature data in the browser. Styled the interface with Bootstrap and ensured responsive behavior for signature capture.

SNAPCLASS AI Attendance System

Built an AI attendance management system that automates student attendance using face recognition in a Streamlit interface. Implemented face detection and recognition using dlib-based tools and added registration and results management screens. Integrated Supabase for secure storage, authentication, QR code generation, and additional audio/image processing utilities to support the AI workflow.

Neural Art

Created a neural style transfer application to generate stylized images by combining content and style from input images. Used CNN-based feature extraction for representing content and style, along with preprocessing and optimization using TensorFlow/Keras. Delivered an end-to-end workflow to transform ordinary images into artistic outputs.

AI Resume ATS Score System

Developed an AI system that evaluates resumes against job requirements and produces compatibility scores. Implemented resume parsing and skill/keyword extraction using NLP with spaCy, including gap analysis. Added an interactive Streamlit workflow for uploading resumes and integrated an LLM for recommendations, with user authentication and persistent storage.

AI Real-Time Gym Coach

Built a real-time fitness coaching app using webcam pose detection to analyze exercise form and performance. Implemented rep and set tracking for multiple exercises and created an interactive dashboard for progress monitoring. Integrated an LLM-based coaching flow with text-to-speech voice feedback and stored workout history in a local database.

Timeline

Chhatrapati Shahu Ji Maharaj University (Kanpur University)

Bachelor's Degree

Kanpur, Uttar Pradesh

Middle AI/ML Engineer Confidence: Medium Generalist

Model Architecture & Training

1/10

How well models are designed and trained

No custom model architectures or training pipelines; the code uses a pretrained MediaPipe pose landmarker and a small notebook with standard sklearn training.

Evidence

AI_REAL_TIME_GYM_COACH/services/vision/exercise_video_processor.py: PoseLandmarker.create_from_options usage

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: training LogisticRegression / LinearSVC / MultinomialNB

Data Pipeline & Feature Engineering

2/10

How data is prepared for models

Basic data preparation and feature engineering: TF-IDF and simple text preprocessing in the notebook, and handcrafted landmark-derived features (angles, drift, balance) for vision detectors.

Evidence

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: preprocess() with tokenization, stemming and TF-IDF

AI_REAL_TIME_GYM_COACH/detectors/biceps_curl.py: process() computes elbow_angle, torso_angle and swing_status from landmarks

Experimentation & Evaluation

3/10

How results are measured and tested

Ad-hoc evaluation exists (train/test split, accuracy, confusion matrix) in a notebook but there is no experiment tracking, reproducible experiment harness, or systematic evaluation pipeline.

Evidence

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: train_test_split, accuracy_score, classification_report

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: confusion_matrix and model comparison plot

MLOps & Deployment

3/10

How models are shipped to production

App-focused deployment using Streamlit and streamlit-webrtc with session state and lightweight persistence; no evidence of production MLOps (CI/CD, model versioning, monitoring).

Evidence

AI_REAL_TIME_GYM_COACH/main.py: webrtc_streamer setup, session_state management and integration with VideoProcessorClass

AI_REAL_TIME_GYM_COACH/services/tracking/metrics.py: sync_metrics_update triggers persistence and voice pipeline hooks

Computational Efficiency

2/10

How efficiently computing resources are used

Some real-time considerations and basic concurrency handling, but no measured GPU optimization, batching, quantization, or profiling evidence.

Evidence

AI_REAL_TIME_GYM_COACH/services/vision/exercise_video_processor.py: threading.Lock for latest metrics and frame timestamping

AI_REAL_TIME_GYM_COACH/main.py: webrtc_streamer called with async_processing=True

Research Depth & Innovation

1/10

Depth of research and new ideas

No research-level innovations or custom algorithmic contributions; detectors are heuristic and notebook uses standard off-the-shelf models.

Evidence

AI_REAL_TIME_GYM_COACH/detectors/biceps_curl.py: heuristic angle-based logic for rep counting and swing detection

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: uses scikit-learn off-the-shelf classifiers without novel modifications

Expertise

AI Agents & Agentic Workflows• Junior

LLM• Junior

Industries

Artificial Intelligence• Middle

Sports• Middle

Technologies

SQL

C++

PostgreSQL

Supabase

Cursor

LangGraph

LangChain

Claude

OpenCV• since 2026

Groq

WebRTC

Computer Vision• since 2026

AI Agents

NLP

spaCy

TensorFlow• since 2026

Keras• since 2026

Git

SQLite

PyTorch

RAG

MediaPipe

TensorFlow C++

Streamlit• since 2026

Pillow• since 2026

CNN• since 2026

OpenAI Codex

GitHub

Text-to-Speech

Machine Learning• since 2026

Machine Learning• mentioned only

Recommendations

Develop real-time fitness coaching features and iterate on the landmark-based detectors into a user-facing product including UI/UX improvements and robustness testing.
Build reproducible training and evaluation pipelines with experiment tracking (W&B or MLflow) and unit/integration tests for detectors and metric computation.
Harden deployment and security: remove hard-coded secrets, add model loading validation, and instrument monitoring (latency, accuracy drift) for the streaming service.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

AI_REAL_TIME_GYM_COACH Student-Performance-Prediction-ML-PROJECT- MCA-PROJECT

Middle Data Scientist Confidence: High ML Practitioner

A practical ML practitioner (approx early-middle level) who builds end-to-end supervised learning prototypes and lightweight LLM agent demos with a focus on classification pipelines. The strongest proven skill is applied supervised modeling and preprocessing shown by end-to-end notebooks that implement TF-IDF preprocessing, label encoding, OneHot encoding, scaling, training multiple classifiers and saving models (credit_wise.ipynb and NAVIOTECH ML PROJECT.ipynb). The public work shows limited evidence of production-grade practices such as cross-validation, hyperparameter tuning, environment pinning, model governance or rigorous business-impact analysis.

Statistical Rigor

2/10

Correct use of statistics

Minimal statistical rigor: models are evaluated with basic metrics but there are no formal assumption checks, uncertainty quantification, significance testing, or robust treatment of multiple comparisons.

Evidence

ML-PROJECT-/credit_wise.ipynb: evaluation prints (precision, recall, f1, accuracy) without confidence intervals or cross-validation

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: model accuracy printed but no statistical tests or uncertainty analysis

Data Wrangling & Cleaning

4/10

Preparing and cleaning data

Reasonable data cleaning and encoding: missing values are imputed, IDs dropped, categorical variables encoded and scaling applied, and simple feature engineering is present.

Evidence

ML-PROJECT-/credit_wise.ipynb: SimpleImputer usage and OneHotEncoder pipeline plus StandardScaler

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: text preprocessing function using NLTK stopwords, tokenization and stemming

Exploratory Analysis & Visualization

3/10

Exploring and visualizing data

Basic EDA and visualization are present but mostly descriptive; plots exist without in-depth written interpretation or hypothesis-driven storytelling.

Evidence

ML-PROJECT-/credit_wise.ipynb: multiple histplots, boxplots and a correlation heatmap

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: confusion matrix and bar chart comparing model accuracies

Predictive Modeling

4/10

Building models that predict

Predictive modeling shows practical end-to-end work with multiple algorithms and evaluation metrics, but lacks robust CV, hyperparameter search, calibration checks, or advanced error analysis.

Evidence

ML-PROJECT-/credit_wise.ipynb: training and evaluation of LogisticRegression, KNeighborsClassifier and GaussianNB with metric reporting

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: TF-IDF vectorization and training of MultinomialNB, LogisticRegression and LinearSVC

Business Insight & Impact

1/10

Turning analysis into business value

Little to no business framing or impact analysis; model evaluation is numeric but does not connect to business costs, decision thresholds, or error-cost tradeoffs.

Evidence

ML-PROJECT-/credit_wise.ipynb: metrics printed for model performance but no discussion of cost of false positives vs false negatives

Reproducibility & Notebook Hygiene

2/10

Clean, repeatable analysis

Some reproducibility basics are present (random_state in splits, model artifacts saved) but environment pinning, packaging, data versioning, and modularization are absent.

Evidence

ML-PROJECT-/credit_wise.ipynb: train_test_split(random_state=42) and pickle.dump used to save models

Student-Performance-Prediction-ML-PROJECT-/NAVIOTECH ML PROJECT.ipynb: pickle.dump of TF-IDF and model objects

Industries

Commerce• Middle

Financial Services• Middle

Technologies

AI/ML

Python• since 2026 • Middle

Scikit-learn

Seaborn

Matplotlib

Pandas

NumPy

LLM• since 2026

NLTK

Machine Learning• mentioned only

Recommendations

Prototype small-to-medium supervised ML solutions that require data cleaning, feature engineering and baseline model comparisons (classification/regression).
Build NLP classification pipelines and lightweight TF-IDF or embedding-based classifiers for document or intent classification tasks.
Develop proof-of-concept LLM-agent integrations and demos that orchestrate simple agent workflows and product recommendation logic.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Student-Performance-Prediction-ML-PROJECT- MCA-PROJECT ML-PROJECT-

Middle Frontend Developer Confidence: Medium UI Engineer

Middle-level UI-focused frontend engineer specializing in interactive, polished vanilla-JavaScript user interfaces and microinteractions. The strongest proven skill is building polished interactive front-end experiences, as shown by the Neural_Art/templates/index.html canvas particle animation, custom range slider bubble logic, and loader overlay. There is little to no public evidence of component-framework architecture, automated tests, backend design, build-system engineering, or measured performance instrumentation.

UI Component Architecture

3/10

How interface parts are built

Modular vanilla-JS patterns and custom UI elements are present but there is no component framework, tokenized design system, or reusable component library.

Evidence

Rohit7824567/DJ_Hack_Residency/app.js: IIFE structure with show(), load(), save() and tab logic organizing the app Rohit7824567/Neural_Art/templates/index.html: custom range slider UI, previewFile() and updateRange() implementing UI behavior

Responsive & Cross-browser

4/10

Works on all screens and browsers

Responsive layout and cross-browser considerations are visible via Bootstrap usage, media queries, and print styles, though advanced feature-detection or RTL/i18n readiness is not shown.

Evidence

Rohit7824567/Neural_Art/templates/index.html: Bootstrap CDN usage and media query for example flows Rohit7824567/DJ_Hack_Residency/style.css: @media print rules and responsive sticky nav/layout rules

Performance Optimization

3/10

Speed of the interface

Runtime performance awareness is shown (canvas using requestAnimationFrame and particle count scaled to width) but no measured performance work, bundle analysis, or advanced code-splitting is present.

Evidence

Rohit7824567/Neural_Art/templates/index.html: animate() uses requestAnimationFrame and numParticles = Math.floor(window.innerWidth / 15) Rohit7824567/Neural_Art/templates/index.html: particle linking logic limits expensive draws by distance checks (d < 100)

Accessibility & Semantics

4/10

Usable for everyone

Accessible patterns appear in places (semantic elements, form labels, aria attributes on accordions) but there is no evidence of automated a11y testing or keyboard focus management for custom widgets.

Evidence

Rohit7824567/Neural_Art/templates/index.html: accordion buttons use aria-expanded, aria-controls and semantic button markup Rohit7824567/DJ_Hack_Residency/index.html: form controls have labels and buttons use clear semantic text

State Management & Data Flow

3/10

Managing data in the app

Client state is handled sensibly with a simple localStorage load/save pattern and input validation, but there are no server-state patterns, request cancellation, optimistic updates, or state machines.

Evidence

Rohit7824567/DJ_Hack_Residency/app.js: load() and save() with try/catch and a KEY-based serialized state object Rohit7824567/Neural_Art/templates/index.html: previewFile() and updateRange() managing ephemeral UI state for uploads and sliders

UX & Visual Polish

5/10

Look and feel quality

High visual polish and thoughtful microinteractions are a clear strength: custom animations, a loading overlay, animated orbs and scanlines, and polished controls and print styles.

Evidence

Rohit7824567/Neural_Art/templates/index.html: loader-overlay, orb CSS, scanlines, and animated range-value bubble Rohit7824567/DJ_Hack_Residency/style.css: stylized slip, animated map dot and print-friendly slip layout

Verified artifacts

mllam/neural-lam• ★ 303 fwitmer/CoastlineExtraction• ★ 31

Expertise

HTML & CSS• Middle

PWA & Web APIs• Middle

Technologies

Frontend

Graphics

JavaScript

Node JS

Express

Bootstrap

React.js

Recommendations

Develop interactive marketing pages and product landing experiences that require bespoke visuals and microinteractions.
Prototype small AI-frontends or demos that need polished file upload previews, sliders, and loading UX.
Build and refine single-page vanilla-JS features where tight control of DOM and animations matters.
Polish print-friendly UIs and small kiosk-style frontends that require careful CSS and print layout.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

Neural_Art DJ_Hack_Residency mini_project-html-css