Jyothika Km
ML Engineer
Python
SQL
JavaScript
Data Pipeline & Feature Engineering: 4/10
MLOps & Deployment: 4/10
Research Depth & Innovation: 4/10
Active 3 days ago
Invite to interview
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
A senior-level ML practitioner specializing in applied computer vision systems and production-facing ML tools with a clear strength in designing end-to-end detection, verification, and dashboarding pipelines. The strongest proven skill is building robust computer vision and verification pipelines that connect detection, multi-metric similarity scoring, and live dashboards used for decisioning. Not evidenced are formal experiment design, systematic statistical validation or calibration, automated tests, and production-scale cloud data pipelines or CI/CD artifacts.
Phone
Technical skills
Languages
3
Python
SQL
JavaScript
Python
4
Flask
SQLAlchemy
FastAPI
Alembic
AI/ML
18
OpenCV
YOLO
NumPy
Pandas
TF-Keras
Streamlit
Pillow
SciPy
PaddleOCR
scikit-image
LLM
Llama
TensorFlow
PyTorch
Whisper
Deep Learning
OpenAI
ResNet
Frontend
3
React.js
Tailwind CSS
Socket.IO
Other
13
Matplotlib
SQLite
PostgreSQL
GitHub
VictoriaMetrics
Vite
Rest API
ESLint
Machine Learning
RAG
Computer Vision
Transfer Learning
Data Augmentation
Timeline
AI/ML Intern
•
Junior
Heftin AI
•
Full-Time
Worked on performance and assessment management for a multi-tenant SaaS platform, defining software architecture, workflows, and data requirements. Implemented backend and database layers using FastAPI, SQLAlchemy, PostgreSQL, and Alembic, including version-controlled migrations with constraints and validation. Conducted structured testing of migration upgrade/downgrade flows and supported an event-driven roadmap for analytics and future AI automation.
Python
FastAPI
SQLAlchemy
PostgreSQL
Alembic
Rest API
Senior AI/ML Engineer
Confidence: High ML Engineer
Computer vision ML engineer at a Senior level specializing in end-to-end vehicle detection, license plate OCR and a multi-modal verification pipeline. The strongest proven skill is applied computer vision and verification algorithm design as implemented in src/modules/vehicle_verification.py which fuses SSIM, color histogram matching, shape descriptors and perceptual hashing into a composite verification result. Public code does not show advanced experiment tracking, large-scale distributed training, or formal test suites and production hardening such as containerization or CI pipelines.
Model Architecture & Training
3/10
How well models are designed and trained
Model construction and training are present but use standard layers and pretrained embeddings without custom training algorithms or large-scale experiment management.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: build_model (hub.KerasLayer + small dense head)
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: main training loop and model.evaluate
Data Pipeline & Feature Engineering
4/10
How data is prepared for models
Solid data engineering and preprocessing across tasks, including dataset splitting, IMDB loading, structured EDA pipelines, and multi-step OCR preprocessing.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: load_datasets splitting using tfds
cost-of-living-analysis/src/analysis.py: load_data, clean_data, compute_indices and merging with GDP
SmartVision_Gate/src/utils/helper.py: preprocess_license_plate multi-step preprocessing pipeline
Experimentation & Evaluation
2/10
How results are measured and tested
Basic experiment and evaluation practices exist - training curves and test evaluation are saved and printed - but no experiment tracking, reproducible runs, ablations, or systematic benchmarking.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: plot_history saving training curves
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: model.evaluate and printed metrics
MLOps & Deployment
4/10
How models are shipped to production
Practical deployment and MLOps attention - model loading in a streamlit dashboard, a Flask API, process orchestration and a dashboard runner with threading and subprocess management.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/dashboard.py: load_model_cached and Streamlit app wiring
SmartVision_Gate/app/app.py: Flask API endpoints and DB initialization
SmartVision_Gate/run.py: run_main_subprocess, create_app and threaded dashboard + subprocess orchestration
Computational Efficiency
2/10
How efficiently computing resources are used
Some efficiency awareness in batching and frame-rate handling, but no measured GPU/quantization/batching optimizations or profiling artifacts.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: use of .batch and configurable batch_size
SmartVision_Gate/run.py: read_video_frames uses FPS-derived delay and loop handling
Research Depth & Innovation
4/10
Depth of research and new ideas
Applied research and algorithmic work is visible in a custom multi-modal vehicle verification engine combining SSIM, HSV histogram matching, Hu moments and pHash with a fused confidence score.
Evidence
SmartVision_Gate/src/modules/vehicle_verification.py: VehicleVerificationEngine class and verify_vehicle method
SmartVision_Gate/src/modules/vehicle_verification.py: compute_structural_similarity, compute_color_histogram_similarity, compute_shape_similarity, compute_perceptual_hash_similarity
Verified artifacts
Expertise
Computer Vision & Image Analysis• Senior
Document Intelligence & OCR• Middle
Industries
Transportation & Logistics• Senior
Technologies
Deep Learning
SQL
PostgreSQL• since 2026
SQLAlchemy• since 2026
VictoriaMetrics
SciPy
Computer Vision• since 2026
Transfer Learning
Llama
TensorFlow
Pandas
PyTorch• since 2026
LLM
RAG
ResNet• since 2026
Whisper
Alembic• since 2026
PaddleOCR
Data Augmentation
OpenAI• since 2026
GitHub
Machine Learning
OCR• mentioned only
Sentiment Analysis• mentioned only
TensorFlow• mentioned only
Recommendations
- Lead development of vehicle-facing CV systems such as LPR pipelines, access-control integrations and verification engines that require robust per-frame decisioning and DB lookups.
- Build and productionize ML-backed web services and dashboards - package the Flask Streamlit apps into Docker images and add CI/CD, health checks and proper secrets management.
- Extend the verification engine with unit/integration tests, benchmark scripts and reproducible experiment tracking (WandB or MLflow) for threshold tuning and ablation studies.
- Harden input handling and model deployment - add model versioning, size/latency budgets, and explicit GPU/quantization experiments if inference efficiency is a priority.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Senior Data Scientist
Confidence: High ML Practitioner
A senior-level ML practitioner specializing in applied computer vision systems and production-facing ML tools with a clear strength in designing end-to-end detection, verification, and dashboarding pipelines. The strongest proven skill is building robust computer vision and verification pipelines that connect detection, multi-metric similarity scoring, and live dashboards used for decisioning. Not evidenced are formal experiment design, systematic statistical validation or calibration, automated tests, and production-scale cloud data pipelines or CI/CD artifacts.
Statistical Rigor
2/10
Correct use of statistics
Limited statistical rigor and uncertainty quantification; thresholds and confidence labels exist, but there are no formal hypothesis tests, cross-validation protocols, calibration, or ablation studies.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: build_model / training loop uses default metrics without CV or calibration
SmartVision_Gate/src/modules/vehicle_verification.py: verify_vehicle uses fixed thresholds and confidence bands but no statistical validation or uncertainty quantification
Data Wrangling & Cleaning
6/10
Preparing and cleaning data
Strong practical data wrangling for both text and images with careful preprocessing, input validation, DB handling, and temp-file cleanup; good attention to leakage prevention for registration flow.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: load_datasets and batching with shuffle
SmartVision_Gate/app/app.py: init_db, register_vehicle (parameterized SQL, duplicate checks, temp file cleanup, sanitize_filename)
SmartVision_Gate/src/utils/helper.py: preprocess_license_plate with multiple enhancement steps
Exploratory Analysis & Visualization
5/10
Exploring and visualizing data
Useful visualizations and dashboards are provided (training curves, Streamlit dashboard, verification visualization) but exploratory analysis is limited to summary metrics rather than deep data-storytelling or documented hypothesis-driven EDA.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: plot_history saves training curves
Classification-of-Movie-Reviews-using-Tensorflow/dashboard.py: analytics and model architecture table
SmartVision_Gate/src/modules/vehicle_verification.py: visualize_verification builds side-by-side comparison and SSIM diff
Predictive Modeling
5/10
Building models that predict
Solid end-to-end predictive work: TF Hub embedding + small classifier and a non-trivial multi-metric vehicle verification pipeline; however, model selection, calibration, CV schemes and error-analysis are minimal.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: build_model, model.fit, predict_texts
SmartVision_Gate/src/modules/vehicle_verification.py: verify_vehicle combining SSIM, color histograms, Hu moments and pHash
Business Insight & Impact
4/10
Turning analysis into business value
Product-focused work connects model outputs to decisions and UX (dashboard, API responses, permission/decision flags), but there is limited explicit business-metric framing or cost-sensitive error analysis.
Evidence
SmartVision_Gate/run.py: SharedState decision/permitted fields and parsing of analyzer output for user-facing reasons
SmartVision_Gate/app/app.py: register_vehicle returns actionable error codes and messages for duplicate/plate mismatch cases
SmartVision_Gate/src/main.py: _build_invalid_reason constructs user-facing failure reasons
Reproducibility & Notebook Hygiene
4/10
Clean, repeatable analysis
Some reproducibility hygiene is present (requirements.txt, argument parsing, model save/load, Streamlit caching) but there is no pinned environment artifact (Docker, lockfile), unit tests, or data/versioning pipeline shown.
Evidence
Classification-of-Movie-Reviews-using-Tensorflow/requirements.txt: pinned package versions
Classification-of-Movie-Reviews-using-Tensorflow/src/train.py: parse_args and model.save to artifacts
Classification-of-Movie-Reviews-using-Tensorflow/dashboard.py: load_model_cached uses streamlit cache_resource
Expertise
Data Science• Senior
Industries
Transportation & Logistics• Senior
Media & Entertainment• Middle
Technologies
Python• since 2023 • Senior
OpenCV
YOLO
Matplotlib
NumPy
SQLite
TF-Keras
Streamlit
Pillow
scikit-image
Analytics• mentioned only
OCR• mentioned only
Sentiment Analysis• mentioned only
TensorFlow• mentioned only
Recommendations
- Develop end-to-end computer vision access-control and vehicle re-identification systems, including verification engines and live dashboards.
- Implement production-ready APIs that integrate ML models with operational databases and simple UX for manual review and exception handling.
- Prototype and harden ML model training pipelines for text and vision tasks with proper experiment tracking, CV, and calibration (add unit tests and reproducible environments).
- Extend verification research into a monitored deployment: add metrics tracking, logging, and automated retraining triggers for drift detection.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Junior Frontend Developer
Confidence: Medium UI Engineer
Junior frontend developer specializing in UI polish and modern project setup with Vite and Tailwind. The strongest proven skill is visual styling and theme polish, supported by src/index.css which implements a dark theme, custom scrollbar, selection styling, and a considered font stack. There is little to no evidence of component architecture, runtime state management, async data handling, accessibility practices, or test coverage in the analyzed source.
UI Component Architecture
1/10
How interface parts are built
Minimal component architecture; analyzed files are static HTML/CSS with no React component code or component boundary decisions present.
Responsive & Cross-browser
2/10
Works on all screens and browsers
Basic responsive and cross-browser considerations exist (viewport meta, min-width, custom scrollbar) but no advanced responsive techniques, container queries, or RTL/i18n readiness.
Performance Optimization
1/10
Speed of the interface
No measured performance work or build-analysis artifacts; only minor client-side UX tweaks (smooth scrolling) are present.
Evidence
Accessibility & Semantics
1/10
Usable for everyone
Minimal accessibility signals (html lang attribute, basic selection styling) but no ARIA, focus management, keyboard handling, or accessibility tooling shown.
State Management & Data Flow
Managing data in the app
Not evidenced in public code
UX & Visual Polish
3/10
Look and feel quality
Clear visual polish and theming (dark background, custom scrollbar, selection color, font stack) indicating a focus on look-and-feel; interactive UX states and complex transitions are not present.
Evidence
Expertise
Frontend Architecture & Build Tools• Junior
Modern Web Frameworks• Junior
Technologies
JavaScript
Tailwind CSS
Socket.IO
React.js
Vite
ESLint
Recommendations
- Build and refine static marketing or portfolio sites using Vite and Tailwind where visual polish and theming are the primary deliverable.
- Convert static designs into accessible, reusable React components and small interactive widgets to demonstrate component architecture and state handling.
- Add basic frontend engineering hygiene: linting and formatting in CI, unit/UI tests, and automated accessibility checks to prove robustness.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: