Prateek Singh
Software Developer
7+ years exp
SQL
JavaScript
Node JS
Ruby
Java
TypeScript
Python
Experimentation & Evaluation: 6/10
Model Architecture & Training: 5/10
Data Pipeline & Feature Engineering: 5/10
Active 2 days ago
+1 (224) 3638696 Invite to interview
Message
Download CVCV
Overview
Technical skills
Timeline
Roles
Overview
A senior-level data-engineering practitioner (senior) focused on reproducible ML-driven benchmarks and analytics pipelines with strong plotting and ETL skills. The strongest proven skill is building end-to-end experiment and analysis pipelines as shown by the RefAgent plotting and modeling stack (java-refactoring-llm-benchmark/notebooks/lib/story_plots.py and notebooks/lib/token_modeling.py). There is less evidence of production MLOps (CI/CD, containerized reproducibility, DVC) and of large-scale distributed data-processing frameworks in public code.
Technical skills
SQL
JavaScript
Node JS• Middle
Ruby
Java• Junior • 7y+
TypeScript• Middle • 6y+
Python• Senior
Node JS
Axios
Java
Maven
Spring Boot• 5y+
Python
Django
Requests
FastAPI
Flask
Gunicorn
Databases
pgvector
PostgreSQL
Qdrant
SQLite
Apache Kafka
DynamoDB
Redis
AI/ML
Claude
Claude Code
CNN
Copilot
CUDA
CUDA Toolkit
Cursor
Flash Attention
LangChain
LangGraph
LlamaIndex
LLM
PEFT
PyTorch
RAG
Reinforcement Learning
SciPy
Streamlit
TensorFlow
Datasets
Tokenizers
TRL
Bitsandbytes
Accelerate
Pandas
NumPy
Scikit-learn
OpenAI SDK
Jupyter Notebook
AI/ML
LoRA
Sentence-Transformers
Spark
Transformers
Frontend
Vite
Tailwind CSS
Zod
React Router
React.js
DevOps
AWS Lambda
Azure
CI/CD
Git
Kubernetes
Rest API
Terraform
Amazon EKS• 5y+
AWS
Grafana
Prometheus
Nginx
Docker
Docker Swarm
Analytics
Matplotlib
Seaborn
Mobile
JUnit
React Native• 7y+
QA
Pytest
Timeline
Graduate Research Engineer
•
Middle
Michigan Technological University
•
Full-Time
Built a production inference service for a fragment-aware molecular property prediction model with a FastAPI backend and React frontend for both batch and interactive molecular queries. Productized a cosmological deep learning simulator as a full-stack web platform on an HPC cluster, using Flask, Gunicorn, and Nginx with SSL termination, rate limiting, async job queuing, checkpoint management, and real-time visualization.
FastAPIsince 2025
React.jssince 2025
Flasksince 2025
Gunicornsince 2025
Nginxsince 2025
Graduate Research Engineer
•
Middle
Michigan Technological University
•
Full-Time
Built an end-to-end inference workflow for a fragment-aware molecular property prediction model, with a FastAPI backend and a React interface for batch and interactive queries. Productionized a deep-learning cosmological simulator as a web platform on an HPC cluster, adding SSL termination, rate limiting, async job handling, checkpointing, and real-time visualization for generated maps.
FastAPI
React.js
Flask
Gunicorn
Nginx
Senior Software Engineer
•
Senior
Capgemini
•
Full-Time
Designed async data pipelines using Kafka (MSK) and Amazon SQS to improve throughput and reduce latency during traffic surges. Implemented multi-tier caching with Redis and DynamoDB to absorb a large portion of requests and maintain low p95 latency. Delivered an observability setup with Prometheus and Grafana, including custom alerting and templated dashboards for consistent service monitoring.
Apache Kafka
AWS
Prometheus
Grafana
Redis
DynamoDB
Software Engineer
•
Middle
Capgemini
•
Full-Time
Developed a Spring Boot microservice for payment and subscription management with banking integrations and SCA flows, supporting millions of users with very high availability. Containerized and deployed multiple microservices on AWS EKS, enabling Blue/Green releases and fast rollbacks. Built an internal attendance platform using geolocation and secure image capture, integrating with HR systems via REST dashboards to reduce audit effort.
Spring Boot
Amazon EKS
Full Stack Developer (Part-time)
•
Middle
KIET Group of Institutions
•
Part-Time
Developed a mobile ERP Android application with student and faculty modules, supporting a large daily active user base. Implemented the app using React Native and Java and connected features through REST APIs. Delivered functionality across client modules with integration to existing backend services.
React Native
Java
Senior AI/ML Engineer
Confidence: High LLM Engineer
LLM engineer (senior-level) specializing in reproducible multi-agent pipelines and benchmark-driven evaluations for code refactoring and model comparison. The strongest proven skill is building and operating a paper-faithful, SLURM-driven benchmarking pipeline and analysis stack, evidenced by refagent/pipeline.py and notebooks/lib/token_modeling.py. Public artifacts do not demonstrate production low-latency serving, formal security audits, or large-scale distributed training engineering details.
Model Architecture & Training
5/10
How well models are designed and trained
Clear, reproducible training experiments and adapter workflows (LoRA/FFT) with concrete training utilities and model loading; solid but not novel architecture research.
Evidence
Mormon-NLT/src/training_utils.py
Mormon-NLT/src/model_utils.py
Mormon-NLT/notebooks/04_exp1_lora_training.ipynb
Data Pipeline & Feature Engineering
5/10
How data is prepared for models
Robust data acquisition and ETL for multiple real datasets, careful derived-table design and verification with row-count checks.
Evidence
SQL-Cheat-Sheet/scripts/fetch.py
SQL-Cheat-Sheet/scripts/build.py
SQL-Cheat-Sheet/scripts/derive.py
Experimentation & Evaluation
6/10
How results are measured and tested
Well-engineered experimentation and evaluation pipelines with reproducible SLURM benchmarks, thorough aggregation, plotting and leave-one-out / bootstrap evaluation.
Evidence
notebooks/archive/03_experiment_results_v0.0.ipynb
notebooks/lib/token_modeling.py
refagent/results_store.py
MLOps & Deployment
5/10
How models are shipped to production
Operationalized LLM inference and jobs orchestration (vLLM/Ollama hooks, model download, SLURM job scripts) with result locking and scripted run procedures.
Evidence
refagent/ollama_utils.py
scripts/run_refagent.py
jobs/download_models.py
Computational Efficiency
4/10
How efficiently computing resources are used
Awareness and usage of efficiency techniques (quantization, AWQ mentions, bitsandbytes, batched RF) but limited low-level profiling or hard optimization artifacts.
Evidence
Mormon-NLT/requirements.txt
notebooks/lib/token_modeling.py
Research Depth & Innovation
4/10
Depth of research and new ideas
Research-minded replication of a paper with paper-faithful prompts, experiment notes and documented limitations, but no clear novel algorithmic contribution.
Evidence
refagent/paper_prompts.py
docs/REPLICATION.md
notebooks/01_overview.ipynb
Expertise
AI / LLM Engineering (Agents)• Senior
AI Agents & Agentic Workflows• Senior
MLOps & Model Lifecycle• Senior
Industries
Artificial Intelligence• Senior
Education• Middle
Software• Senior
Technologies
SQL
PostgreSQL
Redis
Copilot
Cursor
LangGraph
Rest API
LangChain
Claude
Terraform
Spark
pgvector
Flask
DynamoDB
Qdrant
Claude Code
Bitsandbytes
Flash Attention
LlamaIndex
LoRA
Sentence-Transformers
CUDA Toolkit
Docker Swarm
FastAPI
Prometheus
Reinforcement Learning
SciPy
PEFT
Azure
TRL
Accelerate
Tokenizers
Datasets
CI/CD
Transformers
TensorFlow
Django
Git
PyTorch
AWS
Docker
Kubernetes
Nginx
Apache Kafka
Grafana
LLM
RAG
Streamlit
Gunicorn
Requests
Amazon EKS• 5y+
AWS Lambda
CNN
CUDA
Recommendations
- Lead development of multi-agent LLM pipelines and reproducible benchmarking suites (compile/test/metrics loops).
- Build experiment reproducibility and evaluation tooling for LLM-driven code transformations and cost/efficiency studies.
- Implement LoRA/adapter training experiments and adapter deployment flows for constrained hardware targets.
- Create data engineering and teaching products (SQL practice platforms, curated datasets and reproducible ETL pipelines).
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Senior Data Scientist
Confidence: Medium Data Engineer
A senior-level data-engineering practitioner (senior) focused on reproducible ML-driven benchmarks and analytics pipelines with strong plotting and ETL skills. The strongest proven skill is building end-to-end experiment and analysis pipelines as shown by the RefAgent plotting and modeling stack (java-refactoring-llm-benchmark/notebooks/lib/story_plots.py and notebooks/lib/token_modeling.py). There is less evidence of production MLOps (CI/CD, containerized reproducibility, DVC) and of large-scale distributed data-processing frameworks in public code.
Statistical Rigor
6/10
Correct use of statistics
Good applied statistical practice: log-transform models, bootstrap CIs, leave-one-repo-out validation and residual analysis are present, but formal hypothesis testing and multiple-comparison controls are limited.
Evidence
java-refactoring-llm-benchmark/notebooks/lib/token_modeling.py:fit_loglinear and _bootstrap_ci
java-refactoring-llm-benchmark/notebooks/lib/token_modeling.py:leave_one_repo_out
java-refactoring-llm-benchmark/notebooks/archive/03_experiment_results_v0.0.ipynb:Key Findings and SRR caveats
Data Wrangling & Cleaning
7/10
Preparing and cleaning data
Strong, careful data engineering and cleaning: derived tables, type casting, explicit handling of dirty input and reproducible schema extraction; clear ETL scripts for building analytic DBs and project-specific data preparation.
Evidence
SQL-Cheat-Sheet/scripts/derive.py:derived table creation and type casting
SQL-Cheat-Sheet/scripts/build_site.py:build_schema and build_results (PRAGMA inspection, robust query execution)
java-refactoring-llm-benchmark/scripts/lib.py:copy_all_java_src and run_designite (repo-level preprocessing)
Exploratory Analysis & Visualization
7/10
Exploring and visualizing data
High-quality exploratory analysis and narrative visualization with explicit storytelling, colorblind-safe palettes, annotated figures and multiple publication-ready plot types.
Evidence
java-refactoring-llm-benchmark/notebooks/lib/story_plots.py:paper-style plotting, PALETTE, and multiple narrative plot functions
java-refactoring-llm-benchmark/notebooks/archive/03_experiment_results_v0.0.ipynb:story sections, annotated charts and 'Key Findings'
Predictive Modeling
6/10
Building models that predict
Predictive modeling is competent: baseline-first discipline, linear/log-linear and random-forest fits, holdout and leave-one-repo-out validation, and error/residual analysis; models appear research/analysis-focused rather than productionized.
Evidence
java-refactoring-llm-benchmark/notebooks/lib/token_modeling.py:fit_linear, fit_loglinear, fit_rf, evaluate
java-refactoring-llm-benchmark/notebooks/lib/token_modeling.py:run_modeling (train/holdout split, metrics reporting)
Business Insight & Impact
6/10
Turning analysis into business value
Analyses are framed with business questions and impact (cost-benefit, token-efficiency, SRR), and results include actionable metrics and limitations; trade-offs and limitations are explicitly documented.
Evidence
SQL-Cheat-Sheet/queries/05_fraud.sql:Q19 cost-benefit calculation and recommendations
java-refactoring-llm-benchmark/notebooks/archive/03_experiment_results_v0.0.ipynb:Key Findings and limitations section
SQL-Cheat-Sheet/README.md:docs/DATA_NOTES.md referenced caveats and provenance
Reproducibility & Notebook Hygiene
5/10
Clean, repeatable analysis
Reasonable reproducibility hygiene: requirements files, replication docs and seeded analysis paths exist, but there is not a complete pinned environment, CI pipeline, data versioning (DVC) or containerized reproducibility everywhere.
Evidence
java-refactoring-llm-benchmark/requirements.txt:explicit python deps for pipeline and analysis
java-refactoring-llm-benchmark/docs/REPLICATION.md:replication walkthrough
java-refactoring-llm-benchmark/notebooks/lib/token_modeling.py:run_modeling(seed) with reproducible split
Expertise
Analytics• Middle
Industries
Commerce• Middle
Software• Middle
Financial Services• Middle
Technologies
AI/ML
Python• Senior
Jupyter Notebook
Scikit-learn
Seaborn
Matplotlib
OpenAI SDK
Pandas
NumPy
SQLite
Recommendations
- Lead design and implementation of LLM benchmarking and experiment orchestration (SLURM-compatible pipelines, model download and endpoint management).
- Build reproducible analytics pipelines and reporting for research-to-paper workflows, including publication-ready plotting and replication docs.
- Develop data-processing and ETL components that prepare repo-level artifacts for automated model evaluation (class selectors, smell metrics, result JSONL ingestion).
- Implement token-usage and cost-efficiency modeling and tools to estimate inference cost per outcome (extend token_modeling.py into a small library)
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories:
Middle Frontend Developer
Confidence: Medium UI Engineer
Forward-deployed frontend engineer (mid-level) with a strong eye for UI polish and interaction detail. The strongest proven skill is building production-ready client UX and small-scale architecture, evidenced by the custom scroll/key navigation hooks and theme/animation tokens in the site code (Prateeek73/src/hooks and Prateeek73/src/index.css). What is not evidenced is large-scale frontend system design, measured performance tuning artifacts, or a dedicated component design system shared across multiple projects.
UI Component Architecture
6/10
How interface parts are built
Component boundaries and custom hooks show deliberate UI architecture for a personal site, but much of the visual component code is app- or template-style rather than a reusable design system.
Evidence
Prateeek73/src/hooks/useActiveSection.js: custom hook for section tracking, URL and title sync
Prateeek73/src/hooks/useSectionKeys.js: keyboard navigation hook handling focus and typing elements
Prateeek73/src/components/Grid.js: composed presentational React component using CSS tokens
Responsive & Cross-browser
6/10
Works on all screens and browsers
Responsive choices and cross-browser considerations are present (container vs window scrolling, media queries, prefers-reduced-motion), but no advanced fluid/container-query strategies or RTL/i18n scaffolding.
Evidence
Prateeek73/src/index.css: media query that switches scroll container and prefers-reduced-motion rules
Prateeek73/src/hooks/useActiveSection.js: logic choosing #main vs window as scroller depending on computed style
Performance Optimization
5/10
Speed of the interface
Practical performance awareness (throttled scroll handler, passive listeners, preconnect fonts, theme-before-paint) is evident, but there is no evidence of measured optimization artifacts, bundle analysis, or advanced code-splitting.
Evidence
Prateeek73/src/hooks/useActiveSection.js: time-based throttle, passive event listeners
Prateeek73/index.html: preconnect hints and inline theme resolution to avoid flash
Prateeek73/vite.config.js: spa fallback plugin to support GitHub Pages routing (build-time consideration)
Accessibility & Semantics
6/10
Usable for everyone
Good accessibility signals such as keyboard navigation, focus-visible styles and reduced-motion support are present, but there is limited explicit ARIA on custom widgets and no CI a11y tooling shown.
Evidence
Prateeek73/src/hooks/useSectionKeys.js: arrow-key navigation with safeguards for typing elements
Prateeek73/src/index.css: :focus-visible styling and prefers-reduced-motion handling
State Management & Data Flow
4/10
Managing data in the app
Some thoughtful client-side state discipline appears in custom hooks (URL sync, enabled flag) but there is no evidence of advanced async state patterns like request cancellation, optimistic updates or a server-state cache invalidation strategy in the frontend code.
Evidence
Prateeek73/src/hooks/useActiveSection.js: enabled flag and replaceState usage to avoid history pollution
Prateeek73/src/hooks/useSectionKeys.js: coordination with active section and keyboard-driven navigation
UX & Visual Polish
7/10
Look and feel quality
Visual polish and UX details are strong for a personal/site project - theme tokens, animated reveals, scroll snapping and skeleton-like UX considerations are implemented consistently.
Evidence
Prateeek73/src/index.css: design tokens, reveal animations, smooth scroll-snap and accessible motion settings
Prateeek73/index.html: inline script resolving theme before first paint
Expertise
Frontend Architecture & Build Tools• Middle
Web Performance & Optimization• Junior
Industries
Transportation & Logistics• Middle
Technologies
TypeScript• Middle • 6y+
Node JS• Middle
Tailwind CSS
React.js
Vite
Axios
Zod
React Router
Recommendations
- Lead development of polished marketing or portfolio web apps where careful UX and accessibility matter (single-page sites, landing pages, interactive resumes).
- Build small to medium UI libraries or design-token systems that capture the site’s existing tokens and animations for reuse across projects.
- Implement frontend performance audits and create measured before/after artifacts (LCP/INP reports, bundle analysis) to make optimization work explicit.
- Contribute frontend-facing pieces of API integrations (rate-limiting aware clients or auth-token flows) leveraging the backend experience shown in the carrier-integration service.
Repositories
The developer's experience in this domain has been verified based on AI analysis of the following repositories: