Vishwa Prakash
Software Developer
C++
SQL
C
Python
Data Pipeline & Feature Engineering: 5/10
Experimentation & Evaluation: 4/10
Computational Efficiency: 4/10
Active 1 day ago
Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
Handwriting OCR engineer at a Middle level specializing in dataset and I/O engineering for line-level recognition pipelines. The strongest proven skill is building robust data ingestion and preprocessing tooling, evidenced by build_lmdb.py (LMDB cache), IAMDataset (ml/datasets/iam.py), and validate_loader.py's collate/transform validations. There is no human-authored evidence of model architecture design, custom training loops, experiment tracking, or production serving in the analyzed files.
Technical skills
C++
SQL
C
Python• Junior
C++
PyTorch C++
Python
Alembic
bandit
SQLAlchemy
Flask
FastAPI
Databases
SQLite
Firestore
PostgreSQL
AI/ML
Claude
Claude Code
Copilot
Cursor
Gemma
llama.cpp
LLM
Pandas
Phi
PyTorch
Scikit-learn
NumPy
Computer Vision
Data Augmentation
Datasets
DevOps
AWS
Docker
Git
Kubernetes
Ubuntu
Vercel
Debian
Rest API
Docker Compose
CI/CD
Analytics
Matplotlib
QA
Pytest
Frontend
React.js
Timeline
Co-Founder
•
Executive
MealSense
•
Cofounder
Co-built a cross-platform meal-ordering app with a React Native frontend, FastAPI backend, and an isolated Firestore data model per kitchen. Designed a rule-based meal scoring engine using explicit weighted signals as a baseline for later personalization. Added a contextual bandit in shadow mode and evaluated it through an A/B gating approach against the baseline.
React Native
FastAPI
Firestore
Software Developer and Project Lead
•
Lead
Resiligence
•
Full-Time
Architected a B2B synthetic-media detection platform using a microservices approach and designed a central PostgreSQL schema to reduce infrastructure costs. Built a two-vendor detection cascade that runs different moderation flows synchronously and asynchronously. Implemented a versioning mechanism to link corrected evidence back to original records using RFC 3161 timestamps.
PostgreSQL
Santa Clara University
Bachelor's Degree •
Computer Science and Engineering
Junior AI/ML Engineer
Confidence: High Data-centric
Handwriting OCR engineer at a Middle level specializing in dataset and I/O engineering for line-level recognition pipelines. The strongest proven skill is building robust data ingestion and preprocessing tooling, evidenced by build_lmdb.py (LMDB cache), IAMDataset (ml/datasets/iam.py), and validate_loader.py's collate/transform validations. There is no human-authored evidence of model architecture design, custom training loops, experiment tracking, or production serving in the analyzed files.
Model Architecture & Training
1/10
How well models are designed and trained
Minimal evidence for model architecture or training engineering in the human-authored files; only dataset/transform support and validation utilities are present.
Evidence
HandwritingRecognition/ml/validate_loader.py: transform and dataloader checks for expected tensor shapes
HandwritingRecognition/ml/transforms.py: augmentation functions (RandomRotation, ElasticDistortion) used in preprocessing
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Well-implemented data pipeline and preprocessing: dataset loader, LMDB caching, resize logic, and padding collate for batching are present and carefully handled.
Evidence
HandwritingRecognition/ml/datasets/iam.py: IAMDataset implementation with split handling and transform pipeline
HandwritingRecognition/ml/build_lmdb.py: LMDB cache builder with MAP_SIZES, image resizing, and pickled entries
HandwritingRecognition/ml/validate_loader.py: _pad_collate and DataLoader validation (num_workers test)
Experimentation & Evaluation
4/10
How results are measured and tested
Clear validation tooling for data and transforms including automated checks and visual grid output for manual inspection; lightweight but practical evaluation of I/O and transforms.
Evidence
HandwritingRecognition/ml/validate_loader.py: _validate_transforms asserts tensor dtype/shape and _make_grid saves sample PNGs
HandwritingRecognition/ml/validate_loader.py: _validate_dataloader enumerates batches with num_workers=2 to detect worker/pickling issues
MLOps & Deployment
2/10
How models are shipped to production
Some ops useful for production (LMDB cache, CLI scripts) but no model deployment, serving, model versioning or drift monitoring artifacts in the human-authored subset.
Evidence
HandwritingRecognition/ml/build_lmdb.py: CLI to build LMDB cache for fast training I/O
HandwritingRecognition/ml/validate_loader.py: OUT_DIR reporting for diagnostics
Computational Efficiency
4/10
How efficiently computing resources are used
Evidence of efficiency-minded choices at the data layer (LMDB, explicit map_size, custom _ToTensor avoiding numpy C-API issues, padded batching) though no GPU/profiling/quantization work is shown here.
Evidence
HandwritingRecognition/ml/validate_loader.py: _ToTensor uses bytearray + torch.frombuffer to avoid NumPy C-API bridge issues
HandwritingRecognition/ml/build_lmdb.py: MAP_SIZES defaults and buffered PNG writes for I/O performance
Research Depth & Innovation
2/10
Depth of research and new ideas
Augmentation set includes standard and well-documented techniques (elastic distortion, rotation, noise) with reference-level commentary, but no novel algorithmic contributions or reproduced paper experiments in the examined files.
Evidence
HandwritingRecognition/ml/transforms.py: ElasticDistortion implementation (Simard et al. style) and other augmentations
HandwritingRecognition/ml/transforms.py: RandomRotation, GaussianNoise, BrightnessJitter implementations
Expertise
Computer Vision & Image Analysis• Junior
Industries
Artificial Intelligence• Junior
Technologies
SQL
C++
PostgreSQL
Copilot
Cursor
llama.cpp
Claude
Firestore
Claude Code
SQLAlchemy
Vercel
Debian
Gemma
Scikit-learn
Computer Vision
Datasets
Pandas
NumPy
Git
SQLite
PyTorch
AWS
Docker
Ubuntu
LLM
PyTorch C++
Phi
Alembic
bandit
Data Augmentation
Claude• mentioned only
CNN• mentioned only
PyQt• mentioned only
PyTorch• mentioned only
Recommendations
- Implement and document a reproducible training loop and model implementation (e.g., CRNN with CTC) alongside unit-tested training/evaluation scripts to demonstrate model engineering skills.
- Add experiment tracking and reproducibility (W&B or MLflow), with scripted evals on held-out splits and saved run artifacts to show measured model performance.
- Profile and optimize data loading for full training runs: pinned memory, prefetching, batch shuffling behavior, and an evaluation of I/O bottlenecks using realistic loads.
- Expose model export/serving artifacts (TorchScript/ONNX export, a lightweight inference API or Dockerfile and performance benchmarks) to demonstrate deployment readiness.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Intern Mobile Developer
Confidence: Low Generalist
Early-career mobile developer (intern level) with limited original public code available for review and a focus on Flutter-based UIs. The strongest proven capability is producing polished Flutter screens and UI interactions alongside platform integration stubs for audio/voice functionality visible in the public artifacts. There is no verifiable human-authored evidence of advanced mobile concerns such as lifecycle/process-death handling, offline sync engines, measured performance work, background-task orchestration, or release/CI infrastructure.
Platform Native Mastery
Knowing the mobile platform
Not evidenced in public code
Mobile UI/UX & Responsiveness
Smooth mobile experience
Not evidenced in public code
Performance & Battery
Speed and battery use
Not evidenced in public code
Offline & Data Sync
Working offline and syncing
Not evidenced in public code
Device Integration
Using device features
Not evidenced in public code
Release & App Lifecycle
Building and publishing apps
Not evidenced in public code
Recommendations
- Assign to implement and refine Flutter UI screens and flows (permissions, dialogs, navigation) where the developer can iterate quickly and deliver visible progress.
- Pair with a senior engineer to implement platform-channel integrations and complete end-to-end permission flows (rationale, denial handling, settings redirect) on iOS and Android.
- Give ownership of small, well-scoped features such as audio recording/playback, model download with safe partial-file handling, and basic local persistence with migrations.
- Have the developer add unit and widget tests plus a simple CI pipeline (build + test) to demonstrate ownership and raise the evidentiary bar for future evaluations.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Intern DevOps Engineer
Confidence: Low Generalist
Early-career generalist developer focusing on Python applications and basic containerized deployments. Strongest proven skill is simple containerization and deployment automation as shown by MessageHub/Makefile targets and the presence of docker-compose.yml plus a Kubernetes deploy target in the MessageHub README. There is no public evidence of production-grade CI/CD pipelines, rigorous infrastructure-as-code modules, secrets management, or measurable observability and SLO work.
CI/CD Pipelines
Automated build and deploy
Not evidenced in public code
Infrastructure as Code
Managing servers with code
Not evidenced in public code
Containerization & Orchestration
Working with containers
Not evidenced in public code
Observability & Monitoring
Watching system health
Not evidenced in public code
Reliability & Incident Response
Keeping systems up
Not evidenced in public code
Cloud & Cost Optimization
Smart use of the cloud
Not evidenced in public code
Expertise
Platform Engineering & IDP• Intern
Technologies
CI/CD
Python• Junior
Docker Compose
Kubernetes
PyQt• mentioned only
Recommendations
- Own and harden a small microservice deployment: add a GitHub Actions pipeline that builds images, runs unit tests, and deploys to the k8s cluster with a deploy gate and rollback step.
- Replace ad hoc deploy scripts with a repeatable IaC flow - start with parameterized, environment-separated Terraform or Helm charts and add remote state with locking.
- Add secrets discipline and runtime identity: replace any environment-embedded keys with External Secrets or Vault integration and adopt workload identity patterns.
- Instrument the service with observability-as-code: add Prometheus metrics, alerting with basic noise reduction, and structured logs to demonstrate SLO-driven alerts and incident readiness.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: