Mumbai, IndiaSenior Data Analyst @ Iksula
I sit between the
data and the decision.
I build the pipelines, models and dashboards behind a business question — and I'm in the room when the answer has to convince someone.
- Since 2022
- Google via Quilytics · Iksula · Viral Fission
- Sectors
- Retail · Logistics · E-commerce · Consumer marketing
Now
Senior Data Analyst, Iksula
I do the analysis, the data engineering, and the data-science work — including the AI and ML end of it. Alongside that I carry the technical side of pre-sales and solutions: working out what's actually buildable for a client, designing the approach, and being the person who can defend it in the room.
It's the combination I've been moving toward for four years. Analysis that nobody acts on is a hobby; a solution nobody can deliver is a slide. The useful place is in between.
- AnalysisMetrics, modelling, ad-hoc investigation
- Data engineeringPipelines, warehousing, data quality
- AI & MLApplied models, agentic workflows, LLMs
- Pre-sales & solutionsScoping, solution design, technical demos
- 01Demand planning platformMulti-tenant SaaS, forecasting through production planning
- 02Leakage & Recovery AcceleratorContract-breach detection to recovered cash, 7 scenarios
- 03Search Readiness ReviewProduct-data scoring against live SERP and competitor data
Selected work
Client work, mostly. Where a number is quoted it's one I can stand behind; where a chart is illustrative it says so.
01 A demand planning platform that plans past the forecast Iksula · multi-tenant SaaS · built end to end Platform
The problem
Most demand planning tools stop at a forecast and leave the buyer to work out what to actually order. The forecast is the easy half — the hard half is turning it into a purchase order with the right date, the right buffer and a quantity that fits on a truck.
What I did
- Built the platform end to end on FastAPI, React and Docker, multi-tenant from the start, with a BRD and a repeatable client onboarding model.
- Built segmentation and model-zoo backtesting that picks its own champion per series, with ABC/XYZ policies and accuracy/FVA tracking.
- Extended past forecasting into purchase, distribution and production planning — lead-time-derived order dates, buffers sized from measured delivery variability, truck and case rounding, shelf-life gating, fair-share allocation.
- Automated the painful part of onboarding: confidence-scored column auto-mapping, idempotent loads, and a readiness engine that quantifies the forecast-accuracy cost of every data gap.
- Productised a 48-hour Diagnostic — three files in, data readiness and baseline accuracy out — seeding a Diagnostic → Planner → Intelligence ladder.
Outcome
A planning module measured on one honest trust metric: proposals released unedited. A human still issues every indent — the system has to earn the click, not assume it. Onboarding compressed into a 6-week, 8-session model.
Mechanism, not client results — selection runs per series.
- FastAPI
- React
- Docker
- Forecasting
- Backtesting
- ABC/XYZ
- FVA
- Supply planning
- Multi-tenant SaaS
02 Finding the money that quietly leaks out of a contract Iksula · Leakage & Recovery Accelerator · demo → production 7 scenarios
The problem
Contracts say what a fee, a deduction or a payment should be. Reality often disagrees, quietly and at scale. Finance teams find out months later, if at all — and a detection tool that stops at a flagged row has not recovered any cash.
What I did
- Scaled a demo webapp into a shippable product for CFOs and finance teams: re-platformed MongoDB to PostgreSQL, rebuilt the UI on React 19, added CI and AWS ECS Fargate deployment behind a WAF.
- Built a versioned no-code truth layer plus 10 deterministic evaluators that catch fees, deductions and payments breaching signed contract terms — deterministic, so a finding can be defended line by line.
- Broadened the market beyond marketplace partners to 7 leakage scenarios — B2B short payments, overdue receivables, unearned discounts, AP duplicates and overpayments, SaaS overbilling, processing fees, royalty underpayment — packaged as a sellable Scenario Library.
- Closed the loop from detection to recovered cash: escalation, dual-control approval, AI-drafted claim emails, follow-ups, recovery and write-off, with a full audit trail.
- Shipped 6-role access control, so a junior analyst can flag and escalate but never act beyond their authority; multi-brand with a super-admin console.
Outcome
A product that goes all the way from "this invoice breaches the contract" to money actually back in the account — with the paper trail an auditor will ask for.
Write-off is the other terminal state, equally logged.
- PostgreSQL
- React 19
- AWS ECS Fargate
- CI/CD
- Rules engines
- Finance analytics
- RBAC
- Audit trail
03 Proving an SEO problem was a data problem Iksula · enterprise retail client · search readiness review 8.1% vs 19.5%
The problem
A large retailer was ranking for a lot of keywords and converting almost none of that into top-three positions. The instinct in the room was to rebuild the website. That is an expensive instinct to act on without evidence.
What I did
- Scored 500 commercially-weighted products, drawn from a 13,678-product catalogue across 52 sub-categories, on 11 on-page parameters — using a scoring algorithm already validated on 100k+ products.
- Benchmarked against live SERP data — real ranking keywords, real volumes, real positions — plus competitor page crawls, so the comparison was like-for-like rather than keyword guesswork.
- Turned 500 rows into a per-product work queue naming the weakest link and the specific fix, rolled up to the 6 highest-payback fixes in order.
- Delivered 3 Excel studies, a client deck and speaker notes, and scripted the pipeline so it re-runs across the full catalogue and measures movement after fixes land.
Outcome
The constraint came down to one line: 189,696 organic keywords, but only 8.1% of non-branded terms in Google's top 3 — against 19.5% for online-only rivals a fraction of the size. The bottleneck was product data, not site architecture. That reframed the whole remediation budget.
Measured across 189,696 organic keywords.
- SERP data
- Competitor crawls
- Retail analytics
- Scoring models
- Python
- Excel
- Client deck
04 Making Google's logistics dashboards fast enough to act on Quilytics · client project for Google (xWF) · Logistics Analyst ~70% faster
The problem
Google's logistics teams were mid-migration. Dashboards were slow enough that people stopped opening them, and the KPIs on those dashboards had drifted into operational trivia — numbers that moved without telling anyone what to do.
What I did
- Led SQL query optimisation and dashboard engineering on BigQuery, across large-scale logistics datasets.
- Built new logistics metrics — On-Time Receiving (OTR) among them — aligned to real business goals, replacing arbitrary operational KPIs.
- Shipped critical feature requests and metric enhancements so the dashboards matched how teams really worked.
- Turned around ad-hoc analysis and bug fixes fast enough to keep migration timelines intact.
Outcome
Dashboard performance improved by roughly 70%, putting data within near real-time reach. The migration landed smoothly and operations stayed stable afterwards — the part nobody notices when it goes right.
Relative, from the reported ~70% improvement.
- SQL
- BigQuery
- Query optimisation
- Dashboard engineering
- KPI design
- Stakeholder alignment
05 Marketing mix modeling across multiple client accounts Quilytics · Data Analyst Spend strategy
The problem
Clients were allocating marketing budget on last-click attribution and habit. Channels that assisted but didn't close were invisible, so they were systematically underfunded.
What I did
- Built marketing mix models for multiple clients to measure true channel effectiveness.
- Separated genuine channel contribution from baseline and seasonal effects.
- Translated model output into spend recommendations a marketing lead could act on without reading a regression table.
Outcome
Budget conversations moved from opinion to evidence. Clients could see which channels were carrying the result and reallocate accordingly.
Schematic — shows the method, not client results.
- Marketing Mix Modeling
- Regression
- Python
- Time series
- Client reporting
06 Budget pacing that warns you before the money's gone Quilytics · Data Analyst Automation
The problem
Pacing was checked manually and retrospectively. By the time anyone noticed a campaign was overspending or underspending, the window to correct it had closed.
What I did
- Designed and implemented automation frameworks for budget pacing.
- Made spend-versus-target visible in real time rather than at month end.
- Built the alerting so drift surfaced while there was still budget left to steer.
Outcome
Spend control became proactive. The team stopped writing post-mortems about budget and started adjusting mid-flight.
Schematic — shows the mechanism, not client spend.
- Automation
- Python
- Alerting
- Spend analytics
- Real-time reporting
07 Agentic AI solutions, built to actually ship Quilytics · Data Analyst Applied AI
The problem
Plenty of agentic AI demos survive a sales call and nothing after it. The hard part isn't the model — it's the data underneath it and the path to production.
What I did
- Contributed to the design of market-ready agentic AI solutions.
- Built the data pipelines those agents depended on, so they ran on reliable inputs.
- Shaped analytics workflows to be deployment-ready rather than notebook-only.
Outcome
Solutions that held up past the demo — because the pipeline, the data quality and the deployment story were designed in from the start.
Architecture shape, generalised across engagements.
- Agentic AI
- LLMs
- Data pipelines
- Solution design
- Productionisation
08 Turning open-text survey answers into something you can query Viral Fission · Business Intelligence Analyst NLP
The problem
Users answered surveys in free text. That's the richest data in the product and the least usable — thousands of responses nobody could aggregate.
What I did
- Wrote a named entity recognition program in Jupyter to pull structured fields out of open-text responses.
- Added sentiment analysis to classify whether a response was positive or negative about the question asked.
- Built auto-tagging logic so new surveys classified themselves, and used OpenAI to drive campaign-assignment logic.
- Designed the points system scoring Youth Ambassadors and Team Leaders across each IP in the app, and validated the automated allocation algorithm against the intended logic.
- Moved reporting onto BigQuery and GA4, with Tableau on top.
Outcome
Free text became a queryable dimension. Performance scoring stopped being manual, and the dashboards finally reflected what people actually wrote.
Pipeline shape for the survey NLP work.
- NER
- Sentiment analysis
- Python
- BigQuery
- GA4
- Tableau
- OpenAI API
Path
-
Aug 2026 — present
Senior Data Analyst — Pre-Sales & Solutions Iksula
Analysis, data engineering and applied AI/ML, plus the technical side of pre-sales and solutions. Built a multi-tenant demand planning platform (FastAPI, React, Docker) from forecasting to production planning; took a leakage-recovery demo to production on AWS with a no-code contract-truth layer and 10 evaluators across 7 scenarios; and scored 500 products on 11 SEO parameters against live SERP data, isolating product data — not the site — as the constraint.
-
Dec 2024 — Aug 2026
Data Analyst Quilytics
Agentic AI solutions, marketing mix modeling, budget-pacing automation and end-to-end pipelines. Seconded to Google's logistics teams (xWF) as a Logistics Analyst: built new logistics metrics including On-Time Receiving (OTR), and led SQL and BigQuery query optimisation that improved dashboard performance ~70%.
-
Oct 2023 — Nov 2024
Senior Business Intelligence Analyst Viral Fission
Cross-IP user points system driving engagement and retention, survey auto-tagging and OpenAI-driven campaign assignment. Restructured dashboards in SQL, used MongoDB aggregation pipelines to unwind and transform data, and brought GA4 and Google Tag Manager into the lead-to-customer journey.
-
Jul 2022 — Oct 2023
Executive — Business Intelligence Analyst Viral Fission
NER and sentiment analysis on open-text surveys, performance scoring for ambassadors and team leaders, dashboard migration to Angular, and day-to-day lead of the data entry and research interns.
-
May 2022 — Jul 2022
Data Science Intern Forsk Technologies
Data science training and internship programme.
-
Mar 2022 — Apr 2022
Data Analyst Intern Vitara Enterprises
-
Aug 2021 — Oct 2021
Python Instructor Self-employed
Taught Python fundamentals — pandas, NumPy, Matplotlib — to three students at a UK school.
-
Aug 2020 — Sep 2020
Frontend Web Developer XUME
Built the company website front-end with a team of two.
-
Mar 2020 — Apr 2020
Application Analyst XUME
Test cases and QA on the beta app, with a team of three.
Toolkit
Analysis & modelling
- Marketing Mix Modeling
- A/B testing
- Time series forecasting
- Demand planning
- Product analytics
- KPI design
- Statistical analysis
- Retail analytics
- Logistics analytics
Data engineering
- SQL
- BigQuery
- ETL pipelines
- Apache Airflow
- Data warehousing
- Data modelling
- Data architecture
- PostgreSQL
- MongoDB
- Data quality
- Beautiful Soup
AI & machine learning
- Agentic AI
- LLMs
- NLP
- Named entity recognition
- Sentiment analysis
- TensorFlow
- scikit-learn
Visualisation & BI
- Tableau
- Power BI
- Looker Studio
- Google Analytics 4
- Google Tag Manager
- Advanced Excel
- Dashboard engineering
- Query optimisation
Languages & platforms
- Python
- pandas
- NumPy
- R
- Jupyter
- FastAPI
- React
- Docker
- Git
- CI/CD
- Linux
- REST APIs
- AWS (ECS Fargate)
- Azure
- Terraform
- Ansible
Working with people
- Pre-sales solutioning
- Requirement discovery
- Technical demos
- Stakeholder management
- Presentations
- Mentoring
Offline
Before any of this, I competed on skates. National-level skater, and a U19 international medalist for India — one gold, two silver.
Competitive sport teaches a specific thing that transfers: you don't get credit for effort, only for the result on the clock. It's the same instinct I bring to analysis — the work isn't done when the model fits, it's done when someone makes a better decision because of it.
- Gold
- 1
- Silver
- 2
- Level
- U19 International
- Also Strength training and fitness
- Also Building things that don't need to exist
- Also Science fiction, reliably
Credentials
Education
-
BSc, Information Technology2019 — 2022
Bhavan's College, Mumbai · 8.58 CGPA
-
Diploma, Web & Graphics Design2018
E-Connect Computer Education (CBITVT) · A+ · 152/200
-
Higher Secondary, Science2018 — 2019
Nirmala Memorial Foundation College, Mumbai
Certifications
-
BCG GenAI Job SimulationMay 2024
Forage · Python, NLP
-
Complete ML & Data Science BootcampApr 2022
Udemy
-
R ProgrammingApr 2022
Udemy
-
Azure Architect Technologies (AZ-303)Aug 2021
Technical Guftgu
Contact
Got a data problem that's really a decision problem?
Email is the surest way to reach me. I read everything, and I reply to anything with a real question in it.
ronaklodhaps110@gmail.com