Sawan Pandita

Backend • Middle • Management: 1-2 years (1-5 people) • 22 years • State College, USA • UTC-05:00 • In office, Partially remote, Fully remote • Work permits: United States, India • Citizenship: India • English: C2 • Can start now

AI/ML

Management: 1-2 years (1-5 people)

Java

Python

TypeScript

Kotlin

Node JS

C++

JavaScript

SQL

Data Layer & Database: 4/10

Active 3 days ago

Invite to interview

Download CVCV

Overview

Technical skills

Projects

Timeline

Roles

Overview

Backend-focused developer (middle level) with a practical strength in building small-to-medium JVM-based data-backed applications. The strongest proven skill is database-driven application wiring and JDBC-based DAO implementation as shown in StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java which includes table creation, prepared statements, generated keys handling and transaction rollback. There is limited evidence of production-grade API versioning, auth flows, formal migration chains, caches, or distributed systems design in public code.

Technical skills

Languages

Java

Python

TypeScript

Kotlin

Node JS

C++

JavaScript

SQL

Python

FastAPI

Flask

SQLAlchemy

Alembic

AI/ML

LLM

LangChain

PyTorch

BERT

Vertex AI

NumPy

Pandas

DeepSeek

DevOps

GCP

Google GKE

GitLab CI

WebSockets

Kubernetes

Databases

Redis

Apache Kafka

PostgreSQL

Apache Iceberg

Presto

Other

React.js

Axios

Vite

Matplotlib

Pytest

Spark

GitHub

Google Cloud Run

GitHub Actions

SQLite

arXiv

DeepSeek-Coder

Git

GitLab

Rest API

Docker

OpenShift

pySpark

Jupyter Notebook

Model Context Protocol

Machine Learning

CI/CD

Helm

CodeQL

RAG

Deep Learning

AI/ML

Vision

Fine-tuning

Computer Vision

LLM Guardrails

Projects

Real-Time Financial API Integration with Distributed Data Pipelines

Aug 2024 to Dec 2024 4 Months

Built a research prototype streaming market data from a brokerage API to WebSocket clients instead of polling. Implemented resilience to connection drops and API rate limits using automatic reconnection and Redis pub/sub to decouple ingestion from consumers. Targeted low-latency delivery and high update throughput across many tracked symbols.

Bob’s Your Wallet

Team-built a card-rewards assistant for banking relationship managers, including an opportunity scoring approach based on seeded transactions and selected card products. Implemented a FastAPI backend with REST endpoints and a React 19 frontend with many UI components. Integrated IBM Bob and watsonx.ai to support assistant responses and workflows.

Timeline

Penn State University Park (State College)

Bachelor's Degree • Computational Data Science & Mathematics

2023–2027 State College, Pennsylvania

Platform Engineer Co-op • Middle

IBM • Full-Time

Jun 2026 to Aug 2026 2 Months Chicago Partially remote

Built a client-health platform for banking relationship managers with a FastAPI backend and a React frontend, deployed via Docker and Helm to an OpenShift/Kubernetes environment. Integrated an enterprise Dynamics 365 Dataverse API over OData v4 using OAuth2, including robust pagination and retry handling. Added quality gates using pytest suites and Alembic migrations, and implemented AI output validation with an MCP server supplying live app context.

FastAPI

React.js

OpenShift

Kubernetes

Docker

Helm

GitHub Actions

Pytest

Alembic

Model Context Protocol

Software Engineering Intern • Junior

Prolaio • Full-Time

Jun 2025 to Aug 2025 2 Months Chicago In office

Developed a real-time streaming pipeline that broadcasts data changes to downstream consumers using Kotlin with a Kafka sink and idempotent producers. Reduced patient data-entry time by building a patient-facing app backed by a Flask microservice, with a React/TypeScript frontend on Vite deployed on GCP. Improved delivery and QA by automating GitLab CI/CD and adding a Kotlin-based pixel-diff regression tool, collaborating in Agile with teams across multiple European countries.

Kotlin

Apache Kafka

Flask

React.js

TypeScript

Vite

GCP

GitLab CI

Artificial Intelligence Engineering Intern • Junior

Reality AI Lab • Full-Time

May 2024 to Aug 2024 3 Months Partially remote

Served BERT and PyTorch models through a containerized FastAPI service running on GKE. Automated content relevance scoring using an LLM fine-tuning workflow built with LangChain and Vertex AI. Focused on deploying model-serving infrastructure and integrating evaluation/scoring pipelines for AI outputs.

FastAPI

Google GKE

BERT

PyTorch

LangChain

Vertex AI

LLM

Middle Backend Developer Confidence: Medium API Engineer

API Design

2/10

How well APIs are designed

Minimal API design artifacts; code shows controller/DAO separation but no HTTP API contracts, versioning, idempotency keys, or pagination patterns.

Evidence

StudySync-Pro/src/main/java/com/studyplatform/controllers/NotificationController.java: controller layer coordinating DAOs and view refreshes

StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java: DAO interface usage and CRUD methods

Data Layer & Database

4/10

Working with databases

Clear evidence of a working data layer with JDBC usage, prepared statements, generated-key handling and basic transactional handling, but no migration history or explicit schema evolution chain.

Evidence

StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java: createTable, PreparedStatement usage, getGeneratedKeys, insertCourseworks

StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java: deleteStudyPlan uses conn.commit and rollback handling

StudySync-Pro/src/main/java/com/studyplatform/util/DatabaseUtil.java: centralized DB connection utility

Scalability & Performance

2/10

Handling load and speed

Small signals for concurrency and robustness (scheduled executor, request timeout) but no explicit caching, queue-based decoupling, connection-pooling configuration, or measured performance work.

Evidence

StudySync-Pro/src/main/java/com/studyplatform/controllers/NotificationController.java: ScheduledExecutorService used for periodic notifications

ChiLib/ChiLib.py: requests.get(..., timeout=10) indicating basic timeout handling

System Architecture

3/10

Overall system structure

Reasonable module decomposition (models, dao, controllers, views, and a separate CLI module) indicating MVC-style design, but architecture is small-scale and there is no inter-service contract or explicit config/secret management.

Evidence

StudySync-Pro/src/main/java/com/studyplatform/: clear package separation into controllers, dao, models, views

ChiLib/ChiLib.py: modular CLI functions (load_libraries, format_library, display_history) and Config class

Security & Auth

2/10

Protecting data and access

Some input validation and good use of PreparedStatement (reducing SQL injection risk) but no auth/authz, token lifecycle, or documented secrets handling.

Evidence

ChiLib/ChiLib.py: Config.ZIP_REGEX input validation for ZIP codes

StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java: PreparedStatement use for SQL parameterization

Reliability & Observability

3/10

Stability and monitoring

Basic observability and reliability practices present: logging and custom exceptions in the CLI, try-with-resources and rollback handling in DAO, and scheduled background tasks; however retries/backoff, structured tracing, metrics, and graceful shutdown are not evidenced.

Evidence

ChiLib/ChiLib.py: logging.basicConfig setup and custom DataLoadError exception

StudySync-Pro/src/main/java/com/studyplatform/dao/StudyPlanDAO.java: try-with-resources and rollback on failure

StudySync-Pro/src/main/java/com/studyplatform/controllers/NotificationController.java: background scheduler for notifications

Expertise

Java• Middle

Microservices & API Architecture• Middle

Node.js• Junior

Technologies

Java• since 2024 • Middle

Kotlin• since 2025 • Junior

Node JS• Middle

Rest API

Flask• since 2025

FastAPI• since 2024

Apache Kafka• since 2025

Axios

Recommendations

Implement a simple HTTP service wrapper around the Java controllers (e.g., lightweight REST endpoints) to demonstrate API contract/versioning and idempotency handling.
Add a migration history (Flyway or Liquibase) and a small suite of integration tests that run against a real DB to demonstrate schema evolution and test coverage for DAOs.
Introduce connection pooling (HikariCP) and basic metrics (Micrometer) and add retry/backoff for external calls to show operational readiness.
Convert the CLI data loader to a small service with explicit error-retry policy and graceful shutdown to show end-to-end resilience patterns.

Repositories

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

ChiLib PortfolioWebsite StudySync-Pro

Junior AI/ML Engineer Confidence: Medium Generalist

Junior Python ML practitioner focusing on introductory computer vision and data analysis notebooks; the main strength is practical hands-on implementation of tutorial-style PyTorch workflows. The strongest proven skill is building and training a simple CNN end-to-end as shown in notebooks/image-classifier.ipynb where the Net class, training loop, dataset download, and model save are implemented. There is no evidence of production-grade tooling, experiment tracking, advanced model design, deployment pipelines, or reproducible experiment management in public code.

Model Architecture & Training

2/10

How well models are designed and trained

Basic model architecture and training practices are present (simple CNN and training loop) but they follow tutorial patterns with no advanced architecture, regularization study, or rigorous training infrastructure.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: class Net definition and forward implementation sawan18/python_ds/notebooks/image-classifier.ipynb: training loop using optimizer, CrossEntropyLoss and torch.save

Data Pipeline & Feature Engineering

2/10

How data is prepared for models

Basic data handling and preprocessing are implemented (torchvision dataset and transforms, pandas CSV reading) but no robust pipeline, augmentation strategy or leakage controls are evident.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: torchvision.datasets.CIFAR10, transforms.Compose and DataLoader usage sawan18/python_ds/nobel_df.ipynb: pd.read_csv and dataframe exploration patterns

Experimentation & Evaluation

1/10

How results are measured and tested

Minimal experimentation and evaluation: simple accuracy and per-class accuracy reporting are present but there is no validation split, experiment tracking, or reproducible eval pipeline.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: per-class accuracy loop printing class-wise accuracy sawan18/python_ds/notebooks/image-classifier.ipynb: matplotlib usage for visual previews of predictions

MLOps & Deployment

1/10

How models are shipped to production

Tiny amount of MLOps evidence: model state is saved to disk but there is no serving, versioning, CI/CD, or deployment-related code.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: torch.save(net.state_dict(), PATH)

Computational Efficiency

1/10

How efficiently computing resources are used

Minor computational-efficiency considerations exist (DataLoader batching, num_workers), but no GPU/TPU profiling, mixed precision, quantization, or advanced batching/throughput work.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: DataLoader(batch_size=batch_size, num_workers=2) sawan18/python_ds/notebooks/image-classifier.ipynb: use of tqdm for training loop progress

Research Depth & Innovation

1/10

Depth of research and new ideas

No research-depth artifacts or novel algorithmic work; the model mirrors standard tutorial CNN patterns and there are no custom layers or reproduced paper implementations.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: Net class with two Conv2d layers and three Linear layers matching common PyTorch examples

Verified artifacts

jgasperackpsu/march_madness_classify_proj• PR jgasperackpsu/cmpsc448_midterm_project• contrib sayaliph/STAT184ER-FA24_GitHub-Practice• PR schenna217/kong-posh• PR

Expertise

Computer Vision & Image Analysis• Junior

Technologies

AI/ML

SQL• since 2026

C++

PostgreSQL

Redis• since 2024

LangChain• since 2024

Spark

GCP• since 2024

Apache Iceberg

Presto

DeepSeek

Model Context Protocol• since 2026

SQLAlchemy• since 2026

OpenShift• since 2026

Helm• since 2026

GitHub Actions• since 2026

Vertex AI• since 2024

WebSockets• since 2024

Fine-tuning

Computer Vision

GitLab CI• since 2025

CI/CD

Git

SQLite

PyTorch• since 2024

Docker• since 2026

Kubernetes• since 2026

LLM• since 2024

RAG

BERT• since 2024

pySpark

Alembic• since 2026

Google GKE• since 2024

Google Cloud Run

DeepSeek-Coder

GitHub

GitLab

LLM Guardrails

arXiv

Machine Learning

Recommendations

Modularize notebook code into reusable scripts or packages (data loading, model, training, evaluation) so experiments are reproducible and testable.
Add a proper validation split, experiment tracking (W&B, MLflow), and automated logging of metrics and artifacts for reproducible comparisons.
Introduce unit tests and small CI to validate data loading and training code; add clear requirements/environment (e.g., conda env) and a simple run script.
Learn basic deployment and MLOps practices: export model to a standard format, add a minimal inference script or containerized endpoint, and measure latency/throughput.

Repositories

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

python_ds

Junior Data Scientist Confidence: Medium ML Practitioner

A junior ML practitioner focused on hands-on notebook work with basic PyTorch image models and exploratory data analysis. The strongest proven skill is implementing and training a simple CNN with a full train/evaluate loop as shown in sawan18/python_ds/notebooks/image-classifier.ipynb. There is no evidence of production-grade pipelines, systematic validation, testing, or business-impact work in public code.

Statistical Rigor

1/10

Correct use of statistics

Minimal statistical rigor: basic summary statistics and accuracy prints only, no formal tests, uncertainty quantification, or assumptions checks.

Evidence

sawan18/python_ds/nobel_df.ipynb: uses df.describe() and df.info() without formal inference or assumption checks sawan18/python_ds/notebooks/image-classifier.ipynb: prints per-class accuracy and basic evaluation but no confidence intervals or hypothesis tests

Data Wrangling & Cleaning

2/10

Preparing and cleaning data

Basic data loading and selection; simple CSV reads and boolean filtering are present but there is little cleaning, provenance tracking, or leakage prevention.

Evidence

sawan18/python_ds/nobel_df.ipynb: pd.read_csv and boolean selection examples for filtering laureates sawan18/python_ds/notebooks/population.ipynb: pandas.read_csv of data/atlantis.csv and straightforward plotting

Exploratory Analysis & Visualization

3/10

Exploring and visualizing data

Exploratory plots and notebook narrative are present and instructional; visualizations serve teaching and previewing rather than deep, question-driven storytelling.

Evidence

sawan18/python_ds/notebooks/matplotlib.ipynb: multiple pyplot tutorials and example plots sawan18/python_ds/notebooks/image-classifier.ipynb: preview of CIFAR-10 images and plotted model prediction figures

Predictive Modeling

2/10

Building models that predict

Has a basic predictive workflow and a custom PyTorch training loop for a toy image classifier; lacks robust validation schemes, hyperparameter optimization, calibration, and detailed error analysis.

Evidence

sawan18/python_ds/notebooks/image-classifier.ipynb: Net class definition, training loop (optimizer, loss.backward, optimizer.step), model save/load and evaluation

Business Insight & Impact

1/10

Turning analysis into business value

No business framing or impact analysis; there is no discussion of metrics trade-offs, costs of errors, or actionable recommendations tied to stakeholders.

Evidence

sawan18/python_ds/nobel_df.ipynb: exploratory focus without business linkage or cost/error trade-off discussion

Reproducibility & Notebook Hygiene

2/10

Clean, repeatable analysis

Minimal reproducibility: a requirements file and at least one explicit RNG seed exist, but there is no experiment tracking, CI, data/version management, or packaged pipelines.

Evidence

sawan18/python_ds/requirements.txt: lists pinned packages (ipywidgets, matplotlib, pandas, torch, torchvision, tqdm) sawan18/python_ds/nobel_df.ipynb: np.random.seed(0) used for reproducibility within the notebook

Technologies

Deep Learning

Vision

Python• since 2025 • Junior

Jupyter Notebook

Matplotlib

Pandas

NumPy

Recommendations

Build small reproducible experiments: add seeds, documented environment (requirements + lock), simple experiment logs, and a basic train/validate/test split workflow for each notebook.
Prototype baseline models and evaluation pipelines on real or curated datasets, adding cross-validation, simple hyperparameter sweeps, and clearer error analysis.
Convert instructional notebooks into modular scripts or lightweight packages and add unit tests and a README describing how to reproduce key results end-to-end.

Repositories

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

python_ds