Data Science Engineer at Dodge Construction Network

Saurabh Gupta

I build AI systems that get measured, not just demoed.

shipping ML to production: ~5 yearsshipping ML to production
in 5+ ML hackathons: Top 5in 5+ ML hackathons
companies across AI & data: 3companies across AI & data
professional awards: 2professional awards

About

I am a Senior Data Scientist and Applied AI Engineer with ~5 years building and deploying production-grade machine learning and generative AI systems across NLP, computer vision, predictive analytics, and the cloud infrastructure that carries them.

Most of my work is end-to-end ownership of the ML lifecycle: designing custom RAG architectures and deep learning models, then getting them past the notebook and into scalable services on AWS. I build with rigorous data validation, deterministic dependency management, and containerized deployment, which is usually what separates a model that works from a model that keeps working.

The through-line is measurement. I use evaluation evidence to assess whether an AI system is meeting its intended use.

Domains I have worked in

Construction data: LLM-based company trade classification, residential-versus-commercial permit modelling, and real-time scoring services over commercial construction and building-permit records.
Supply chain: Data quality automation and partner-data ingestion pipelines for a global beverage manufacturer, including schema validation, reconciliation and anomaly detection across 20+ datasets under delivery SLAs agreed with external stakeholders.
Mobility & parking: Vehicle re-identification from entry and exit imagery, enforcement routing and illegal-parking revenue prediction across urban parking networks in Norway, Denmark, Germany, and Sweden.
Master data management: Entity resolution, crosswalk-safe remediation and large-scale data quality automation in Reltio, recovering missing organization records and purging invalid contacts without losing cross-source lineage.

Open to

Senior Data Scientist, Machine Learning Engineer, AI / GenAI Engineer, MLOps Engineer, currently based in Bengaluru, Karnataka; open to relocating worldwide.

Skills & tooling

PRODUCTION AIframe → model → evaluate → ship

5 branches · 43 capabilities

Core competencies: Generative & Agentic AI
LLM Orchestration
Retrieval-Augmented Generation
MLOps
Natural Language Processing
Computer Vision
Languages: Python
SQL
Shell scripting
Libraries & frameworks: LangChain
LangGraph
Ollama
FastAPI
Pydantic
PyTorch
TensorFlow
Scikit-Learn
Hugging Face
Cloud & infrastructure: AWS ECS
AWS Fargate
SageMaker
Bedrock
AWS Batch
Glue
ECR
RDS
Docker
Terraform
GitHub Actions
MLflow
DVC
PostgreSQL
Alembic
Machine & deep learning: Linear & Logistic Regression
Decision Trees
Random Forest
XGBoost
K-Means
CNNs
Siamese Networks
Transfer Learning
TF-IDF
BM25

Experience

Dodge Construction Network

Bengaluru, Karnataka

full time

Data Science Engineer

Jul 2025 - Present

1 yr 3 mos

Role scope

Owned applied AI delivery across construction-data products, from problem framing and model development to production deployment, data enrichment, and operational reliability.

Selected work

2 of 6

Trade classification engine
Generative AI
Engineered an LLM-powered company trade classification pipeline on LangChain, LangGraph and AWS Bedrock.
classification accuracy
85%classification accuracy
less manual processing
70%less manual processing
Production AI services
AI Platform
Built and deployed real-time AI backend services on FastAPI, Docker and AWS ECS/Fargate, with rate limiting, elastic auto-scaling, strict Pydantic validation and Alembic-managed database migrations.

Systems & tools

LangChain
LangGraph
AWS Bedrock
FastAPI
Pydantic
Docker
AWS ECS
Fargate
AWS Batch
Terraform
MLflow
DVC
PostgreSQL
Alembic
Reltio MDM

Sigmoid Analytics

Bengaluru, Karnataka

full time

Senior Data Scientist

May 2025 - Jul 2025

2 mos

Role scope

Delivered data-quality and ingestion solutions for partner datasets, translating reliability requirements into repeatable checks, workflows, and stakeholder-ready insights.

Selected work

2 entries

Data-quality automation
Data Quality
Automated data quality workflows including schema validation, null checks, reconciliation, and anomaly detection, and established delivery SLAs with external stakeholders.
datasets covered
20+datasets covered
Partner ingestion validation
Data Engineering
Developed scalable validation pipelines for partner data ingestion, improving reliability and refresh consistency.
less manual effort
40%less manual effort

Systems & tools

Python
SQL
Data Validation
Anomaly Detection

Get My Parking

Bengaluru, Karnataka

full time

Associate Data Scientist

Jan 2022 - May 2025

3 yrs 4 mos

Role scope

Built data products for parking operations, combining vehicle intelligence, geospatial analysis, and predictive modelling to improve field decisions and platform reliability.

Selected work

2 of 10

Vehicle image matching
Computer Vision
Built an AI-powered vehicle image similarity system to match vehicles at entry and exit, mitigating errors from License Plate Recognition cameras. Used a Siamese network with contrastive learning and transfer learning on a ResNet backbone.
top-3 match accuracy
87%top-3 match accuracy
Low-light enhancement
Computer Vision
Built a low-light image enhancement pipeline with Zero-DCE to recover vehicle details in challenging illumination and improve the quality of images passed to the similarity model.

Data Science Intern

Jul 2021 - Jan 2022

6 mos

Role scope

Supported digital route-planning research by preparing external and internal data, testing assumptions, and turning exploratory analysis into actionable model inputs.

Selected work

3 entries

Route-planning research
Applied Research
Worked with the team testing Digital Route Planning (DRP), covering the full project pipeline from collection to analysis.
External-data acquisition
Data Engineering
Wrote web scraping scripts to collect weather and event data using requests, Scrapy and BeautifulSoup, and extracted raw data from MySQL servers.
Parking-pattern validation
Statistical Analysis
Performed hypothesis testing to validate whether weather and event conditions affect city parking patterns, helping the team assess model performance.

Systems & tools

PyTorch
Siamese Networks
ResNet
YOLOv8
Zero-DCE
XGBoost
Scikit-Learn
KMeans
Python
MySQL
Scrapy
Power BI

Projects

Things I built outside of work, with evaluation details where available. Source links are included where available.

Generative AI
Vernacular RAG Voicebot for Government Schemes
A public-prototype assistant for government-scheme questions in Hindi and Bengali voice or English, Hindi, and Bengali text, with a grounded RAG workflow and local-by-default inference.
English context recall · offline
0.97English context recall · offline
English context precision · offline
0.91English context precision · offline
English faithfulness · answered rows
0.996English faithfulness · answered rows
- LangGraph
- Ollama
- Gemma
- AI4Bharat Conformer
- Parler-TTS
- BM25
- Cross-Encoder Reranking
- RAGAS
- Docker
- FastAPI
View case study: Vernacular RAG Voicebot for Government Schemes
View source
Machine learning
MSME Financial Health Card
An explainable credit-workflow prototype scored on synthetic snapshots of alternative-data fields, keeping missing repayment history explicit while scoring the remaining evidence.
eligibility ROC-AUC · synthetic holdout
0.978eligibility ROC-AUC · synthetic holdout
generated-PD adj. R² · synthetic holdout
0.905generated-PD adj. R² · synthetic holdout
approval spread · synthetic portfolio
2.4 ptsapproval spread · synthetic portfolio
- Python
- Scikit-Learn
- XGBoost
- FastAPI
- Pandas
- Credit Risk Modeling
View case study: MSME Financial Health Card
View source

Credentials

Honors & awards

Value Evangelist - Grow Together Award, Get My Parking, 2024
Open original
2024
Value Evangelist - Grow Together Award
Get My Parking
For the deployment and enhancement of Digital Route Planning applications across Denmark, Norway and Sweden.
Go Getter Award, Get My Parking, 2023
Open original
2023
Go Getter Award
Get My Parking
For automating reporting workflows and generating operational insights for the business.

Certifications

Neural Networks and Deep Learning certificate, DeepLearning.AI
Open original
DeepLearning.AI
Neural Networks and Deep Learning
Machine Learning Masters certificate, Coding Blocks
Open original
Coding Blocks
Machine Learning Masters
Docker for Machine Learning certificate, CampusX
Open original
CampusX
Docker for Machine Learning
GenAI using Ollama certificate, CampusX
Open original
CampusX
GenAI using Ollama
HackerRank Advanced SQL certificate, Hackerrank
Open original
Hackerrank
HackerRank Advanced SQL

Competitions

Rank 1 of 1600+ participants, HackerEarth Machine Learning Challenge, 2025
Open original
2025
Rank 1 of 1600+ participants
HackerEarth Machine Learning Challenge
First place across the full participant field in a competitive machine learning challenge.
1st Runner Up amongst 12+ teams, Get My Parking, 2024
Open original
2024
1st Runner Up amongst 12+ teams
Get My Parking
For creating a car and number plate matching image similarity model by combining SOTA computer vision techniques like CLIP (Contrastive Language-Image Pre-training) by OpenAI and fuzzy matching methods like Levenshtein distance to accelerate the review case resolution for ExpressAI product.
Ranked 4/342 participants, Dockship, 2021
Open original
2021
Ranked 4/342 participants
Dockship
Fourth place across the full participant field in a competitive machine learning challenge.
Ranked 20/2300+ participants, HackerEarth, 2020
Open original
2020
Ranked 20/2300+ participants
HackerEarth
Achieved 20th place across the full participant field in a competitive machine learning challenge.

Education

2017-2021
Lovely Professional University
Bachelor of Technology (Hons.) · Computer Science Engineering
CGPA 8.36/10
Punjab, India

Get in touch

Open to thoughtful collaborations

Have a hard AI problem worth measuring?

If you are taking an ML system beyond the notebook, tightening evaluation around a GenAI product, or exploring a technical collaboration, I would be glad to hear what you are working on.

Best place to start

Put the problem on the table.

A short note about the context, the constraint, and what a useful outcome looks like is plenty. Email is the most reliable way to reach me.

Direct email

saurabh48782@gmail.com

Start a conversationBengaluru, Karnataka · open to working across time zones

Good reasons to reach out

01
Build something together
Production AI, retrieval systems, evaluation pipelines, or an applied ML idea with a real user and a clear measure of success.
02
Pressure-test an idea
Compare architecture choices, experiment design, or how to tell whether a model is actually improving the product around it.
03
Share the work
Open-source experiments, hackathons, technical writing, or a thoughtful conversation with people working on similar problems.

Elsewhere

Saurabh Gupta

About

Domains I have worked in

Open to

Skills & tooling

Experience

Trade classification engine

Production AI services

Data-quality automation

Partner ingestion validation

Vehicle image matching

Low-light enhancement

Route-planning research

External-data acquisition

Parking-pattern validation

Projects

Vernacular RAG Voicebot for Government Schemes

MSME Financial Health Card

Credentials

Honors & awards

Certifications

Competitions

Education

Get in touch

Have a hard AI problem worth measuring?

Put the problem on the table.

Build something together

Pressure-test an idea

Share the work