Asikur Rahaman
NLP Researcher
C++
JavaScript
C
Java
Python
SQL
Data Pipeline & Feature Engineering: 5/10
Experimentation & Evaluation: 5/10
Model Architecture & Training: 4/10
Active 2 days ago
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
Document-intelligence ML engineer (middle level) with practical experience building CLIP-based visual embedding pipelines and KNN/MLP classifiers for handwritten-prescription mapping. The strongest proven skill is end-to-end vision pipeline engineering using frozen CLIP embeddings plus a clear evaluation/thresholding flow, demonstrated in the Demo2.ipynb notebook (embedding computation, KNN baseline, threshold-grid confidence selection). Public code lacks production-grade MLOps, automated testing, privacy safeguards and any evidence of large-scale deployment or advanced model optimization techniques.
Technical skills
C++
JavaScript
C
Java• Junior • 3y+
Python• Middle
SQL• Junior
Databases
MySQL
AI/ML
EasyOCR
Embeddings
Google Colab
Streamlit
ChatGPT
Deep Learning
Transformers
OpenCV
Scikit-learn
NumPy
Pandas
Torchvision
CLIP
Computer Vision
Frontend
daisyUI
Tailwind CSS
PostCSS
React.js
Vite
ESLint
React Router
DevOps
Git
Analytics
Matplotlib
Plotly
Power BI
Seaborn
Timeline
East West University
Bachelor's Degree •
Data Science
Middle AI/ML Engineer
Confidence: High ML Engineer
Document-intelligence ML engineer (middle level) with practical experience building CLIP-based visual embedding pipelines and KNN/MLP classifiers for handwritten-prescription mapping. The strongest proven skill is end-to-end vision pipeline engineering using frozen CLIP embeddings plus a clear evaluation/thresholding flow, demonstrated in the Demo2.ipynb notebook (embedding computation, KNN baseline, threshold-grid confidence selection). Public code lacks production-grade MLOps, automated testing, privacy safeguards and any evidence of large-scale deployment or advanced model optimization techniques.
Model Architecture & Training
4/10
How well models are designed and trained
Practical model-building with clear PyTorch model classes, a training loop, and reasonable architectural choices (frozen CLIP + small MLP head); not research-novel and no extensive custom layers or advanced training schedules.
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: class MultiInputCLIPClassifier (definition and forward)
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: class VisualOnlyCLIPClassifier (definition and forward)
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: training loop using optimizer=AdamW, CrossEntropyLoss, model.classifier.parameters()
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Solid end-to-end data pipeline and feature work: filename mapping, per-class holdout split, OCR caching, header cropping, embedding computation and a sensible KNN baseline on frozen embeddings.
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: find_file_in_images and filename mapping into df['filename']
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: OCR caching flow with easyocr.Reader and ocr_cache.json handling
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: per-class holdout split cell and compute_emb_for_df function
Experimentation & Evaluation
5/10
How results are measured and tested
Conscientious experimentation and evaluation: KNN baseline vs classifier, top-1/top-3 metrics, confusion analysis, threshold-grid for confident-auto-accept, and ablation (visual-only). No experiment tracking service or reproducible pipelines noted.
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: KNN baseline computation (compute_emb_for_df then KNeighborsClassifier fit and metrics)
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: threshold grid evaluation cell computing coverage vs confident_acc
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: inference + top-1 & top-3 evaluation and confusion/pred-freq analysis
MLOps & Deployment
2/10
How models are shipped to production
Basic model persistence and artifacts saved (torch.save, joblib.dump, excel/csv outputs) but no evidence of serving, CI/CD, model versioning, monitoring or deployment pipelines.
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: torch.save({'model_state': ...}, os.path.join(MODELS_DIR, ...))
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: joblib.dump(knn, KNN_PATH) and writing outputs to /outputs/*.xlsx / .csv
Computational Efficiency
2/10
How efficiently computing resources are used
Minimal efficiency work: reasonable use of batching and DataLoader, KNN with n_jobs=-1, and GPU use where available, but no profiling, quantization, or multi-GPU/distributed optimization.
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: BATCH_SIZE, DataLoader usage in training loop
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: KNeighborsClassifier(n_jobs=-1) and device selection 'cuda' if available
Research Depth & Innovation
2/10
Depth of research and new ideas
Applied experimentation and ablation but no novel algorithms, no paper-reproduction fidelity checks, and no advanced research contributions; mostly applied assembly of known components (CLIP + KNN/MLP).
Evidence
Automated-Mapping-of-Handwritten-Prescription-Images/README.md: design summary describing CLIP embeddings + KNN (matches implemented pipeline)
Automated-Mapping-of-Handwritten-Prescription-Images/Demo2.ipynb: ablation cells comparing visual-only vs text+visual
Expertise
Document Intelligence & OCR• Middle
Computer Vision & Image Analysis• Middle
Industries
Health Care• Middle
Education• Junior
Technologies
Deep Learning
Python• Middle
SQL• Junior
C++
MySQL
ChatGPT
OpenCV
Embeddings
Scikit-learn
Computer Vision
Google Colab
Transformers
Pandas
NumPy
Git
Streamlit
CLIP
Torchvision
EasyOCR
Recommendations
- Productize the pipeline: add a lightweight inference service (REST/gRPC) and model versioning plus unit/integration tests to make the system deployable.
- Harden data practices and privacy: add explicit PHI handling, anonymization or consent notes and dataset provenance, and include a model card / datasheet for transparency.
- Introduce experiment tracking and reproducibility: use W&B/MLflow or structured run scripts and seed-controlled training to record hyperparameters and metrics.
- Improve efficiency and robustness: add batching/async preprocessing for OCR and embeddings, consider FAISS for nearest neighbors at scale, and add simple profiling to find hotspots.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Junior Backend Developer
Confidence: Medium Generalist
Entry-level Java developer (junior) focused on basic object-oriented programming and console-based applications. The strongest proven skill is class design and constructor chaining demonstrated in Teacher.java and Manager.java. There is no public evidence of backend work such as APIs, databases, testing, CI/CD or production-ready system design.
API Design
How well APIs are designed
Not evidenced in public code
Data Layer & Database
Working with databases
Not evidenced in public code
Scalability & Performance
Handling load and speed
Not evidenced in public code
System Architecture
Overall system structure
Not evidenced in public code
Security & Auth
Protecting data and access
Not evidenced in public code
Reliability & Observability
Stability and monitoring
Not evidenced in public code
Expertise
Java• Junior
Technologies
Java• Junior • 3y+
Recommendations
- Develop small REST APIs using Spring Boot to demonstrate API design, routing, error contracts and idempotency handling.
- Add unit and integration tests and a simple CI pipeline to show test coverage and delivery discipline.
- Integrate a relational database and provide migrations to demonstrate schema evolution and transaction handling.
- Build one service with observability (structured logs, basic metrics) and clear timeout/retry handling to show production-readiness.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Junior Frontend Developer
Confidence: Low UI Engineer
A junior Frontend UI engineer focused on responsive HTML/CSS and simple React/Vite projects. The strongest proven skill is building responsive, visually polished static interfaces as shown in Assignment2_ph/index.html and Assignment2_ph/styles/style.css. There is little public evidence of complex state management, automated tests, accessibility engineering beyond basic semantics, or production-grade React component architecture.
UI Component Architecture
2/10
How interface parts are built
Very limited component architecture; no authored React components or component boundaries are present and the project is mostly static pages with a basic Vite setup.
Evidence
Accountancy-project/vite.config.js: Vite + @vitejs/plugin-react plugin present but no src/components folder or JSX files shown
Accountancy-project/src/index.css: project CSS present but no React component CSS modules or component-specific structure
Responsive & Cross-browser
3/10
Works on all screens and browsers
Responsive layouts implemented by hand in CSS using grid/flex and a small media query set; pragmatic but not advanced (no container queries, no RTL/i18n patterns).
Evidence
Assignment2_ph/styles/style.css: @media screen and (max-width:576px) responsive rules and grid/flex layout patterns
Assignment1_ph/style.css: use of flex layouts and background-size adjustments for different view widths
Performance Optimization
1/10
Speed of the interface
No measurable performance work or optimization artifacts; only basic modern tool usage (Vite / Tailwind plugin) is present.
Evidence
Accountancy-project/vite.config.js: uses @vitejs/plugin-react and @tailwindcss/vite plugin but no bundle analysis or performance audit artifacts
Accessibility & Semantics
2/10
Usable for everyone
Basic semantic HTML and image alt attributes exist; no ARIA, focus management, keyboard handling or CI a11y tooling evidence.
Evidence
Assignment2_ph/index.html: lang attribute, meta viewport and alt attributes on images
Assignment1_ph/index.html: alt attributes on images and semantic headings
State Management & Data Flow
2/10
Managing data in the app
Minimal server/state discipline; a Firebase init shows awareness of env-based configuration but there are no state machines, optimistic updates, request cancellation or non-trivial data flows.
Evidence
Accountancy-project/src/Firebase/firebase.init.js: initializeApp with env-based config (import.meta.env.VITE_...)
UX & Visual Polish
3/10
Look and feel quality
Good visual polish for static pages with deliberate typography, card layouts and responsive adjustments, but UX patterns like skeletons, undo flows or progressive loading are not present.
Evidence
Assignment2_ph/styles/style.css: detailed styling for banners, cards, buttons and responsive rules
Assignment2_ph/index.html: structured content (hero, features, team, BMI calculator UI) demonstrating visual composition
Expertise
Modern Web Frameworks• Junior
HTML & CSS• Junior
Frontend Architecture & Build Tools• Junior
Technologies
JavaScript
Tailwind CSS
React.js
Vite
PostCSS
ESLint
daisyUI
React Router
Recommendations
- Build marketing or informational websites and landing pages where responsive HTML/CSS and visual polish are primary requirements.
- Implement small Vite + React apps with Firebase-backed auth/CRUD to gain experience with component design and server-state patterns.
- Develop a small design-system or a set of reusable UI components (buttons, cards, form controls) to practice component boundaries, accessibility, and state isolation.
- Add basic frontend testing (Jest/React Testing Library) and simple performance checks (Lighthouse, bundle analysis) to demonstrate engineering maturity
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: