S
Solvarya

Technical Proof of Capability

Technical Work & Projects

An auditable archive of independent projects, prototypes, and data engineering implementations demonstrating Solvarya's technical capabilities. These are explicitly personal implementations and research prototypes, not completed client engagements. All datasets and limitations are disclosed transparently.

Fulfillment Risk & Delay Engine100k Orders Analyzed
Delay Rate
13.24%
Perishables 18.1%
Delayed Refund
46.8%
3.2× On-Time
Model ROC-AUC
0.839
vs 0.828 Baseline
Stratified 80/20 Split (5-Fold CV)Streamlit + SQL CTEs
Analytics & Business Intelligence
Completed Personal Project
Synthetic Dataset | 100,000 Simulated Orders
Primary Commercial Analytics Showcase

E-Commerce Order Fulfillment Operations Analysis

An end-to-end operations analytics pipeline examining delivery delays, refund patterns, category performance, and order-level refund risk using a reproducible synthetic dataset.

Business Problem

E-commerce platforms frequently encounter delivery delays, perishable fulfillment bottlenecks, and escalating refund requests. Without order-level analysis, operations teams cannot isolate whether customer complaints stem from courier SLA breaches, category handling issues, or service ratings.

Analytical Approach

Seeded synthetic generator with 21 validation checks, ETL cleaning with Pandas, PostgreSQL analytical queries using CTEs and window functions, stratified 80/20 train/test split with 5-fold cross-validation, and an interactive Streamlit UI.

Key Findings

•Analyzed 100,000 simulated orders and 2,302 simulated customers across synthetic multi-platform contexts

•Identified overall delay rate of 13.24%, with Fruits & Vegetables highest (18.14%) and Grocery most reliable (9.51%)

Actionable Recommendations

•Implement courier SLA breach monitoring to identify fulfillment delays before customer refund requests escalate.

•Prioritize dispatch queues for perishable categories (18.14% synthetic delay rate) to mitigate high refund exposure.

•Use an explainable rule-based SQL score (0.828 ROC-AUC) as an initial baseline before deploying trained machine learning models.

Data Source TransparencyUses a synthetic, reproducible dataset of 100,000 simulated orders generated by a seeded script (validate_data.py) with 21 data-validation checks. Platform names represent simulated business contexts for analytical modeling, not actual business performance of external entities. Findings illustrate statistical patterns within synthetic test data; correlation does not imply causation.
PythonSQLPostgreSQLPandasScikit-learnStreamlit
SmartOps Multi-Service Core99,441 Olist Orders
Fulfillment
97.1%
Delivery Rate
At-Risk Churn
41.3%
RFM Clustering
Flagged Sellers
149
Isolation Forest
FastAPI + Pydantic RESTChromaDB RAG (<45ms)
AI & Applied Intelligence
Demonstration Prototype
Public Brazilian Olist Dataset | 99,441 Orders
Applied AI & Engineering Prototype

AI-Powered E-Commerce Operations Intelligence System (SmartOps AI)

An integrated operational intelligence prototype combining demand forecasting, customer RFM segmentation, seller anomaly detection, and document-grounded policy retrieval.

Business Problem

E-commerce merchants frequently manage fragmented operational data: internal return policies and shipping guidelines live in disconnected documents, while demand forecasts and anomaly detection scripts remain isolated in offline notebooks without standard APIs for operations teams.

Analytical Approach

Decoupled microservice architecture with FastAPI/Pydantic, Prophet time-series demand modeling, RFM customer clustering, Isolation Forest anomaly scoring, and ChromaDB semantic vector search with strict prompt bounding.

Key Findings

•Evaluated on 99,441 real public orders spanning 25 months totaling R$20.4M in GMV from the Brazilian Olist dataset

•Measured 97.1% overall delivery fulfillment rate across Brazilian logistics routes

Actionable Recommendations

•Use RFM behavioral segmentation to identify the 41.3% of customers in the At-Risk cluster for targeted re-engagement campaigns.

•Establish automated delay alert thresholds for the 149 merchants exhibiting multi-week fulfillment latency anomalies.

•Enforce strict context boundaries in document retrieval prompts so that missing documentation triggers structured fallbacks rather than ungrounded completions.

Data Source TransparencyDemonstration prototype evaluated on the public Brazilian Olist e-commerce dataset containing 99,441 real orders across 25 months (2016–2018) totaling R$20.4M in GMV. This is an engineering research prototype, not a production client deployment.
PythonFastAPIPandasChromaDBProphetScikit-learnStreamlitDocker
RFM Customer Segmentation200k+ Records
Champions
24.2%
Loyalists
38.5%
At-Risk
14.1%
Star Schema (SQL)Power BI Ready
Analytics & Business Intelligence
Completed Personal Project
Public Online Retail Dataset

E-Commerce Analytics & Customer Segmentation

An end-to-end data analysis pipeline evaluating transaction history, customer retention curves, and RFM segments with interactive Power BI dashboards.

Business Problem

Raw multi-category e-commerce transaction logs frequently contain data anomalies, cancelled orders, and unstandardized dates, preventing management from gaining clear visibility into repeat customer lifetime value and retention trends.

Analytical Approach

Python data cleaning with Pandas to handle cancellations and returns, star-schema modeling in PostgreSQL with dedicated fact and dimension tables, monthly cohort retention analysis, and dynamic DAX metrics in Power BI.

Key Findings

•Cleaned raw transaction logs handling missing customer identifiers, cancellations, and negative returns

•Constructed relational star schema with fact and dimension tables optimized for analytical queries

Actionable Recommendations

•Address negative return quantities and missing customer IDs during initial ETL to protect downstream retention metric validity.

•Segment marketing communications across RFM tiers (Champions, Loyalists, At-Risk, Hibernating, New) to improve customer retention.

Data Source TransparencyDemonstration analytics project evaluated on the public Online Retail transaction dataset. All data cleaning, cohort modeling, and dashboard metrics reflect open research data, not proprietary commercial client records.
PythonPandasSQLPostgreSQLPower BIDAXExcel

Get in Touch

Have a Project in Mind? Let’s Talk.

Tell me about your agency challenge, spreadsheet bottlenecks, or workflow needs. I will review the details and suggest a practical approach.

Direct Communication Channels

Prefer direct communication over the form? Send an email or connect via WhatsApp:

vinay.opsanalytics@gmail.com
Response within 24 business hours
Confidential scoping & no sales spam