Sanskriti Rai · Cloud & Data Architect

Business brain. Architect's hands. AI at scale.

I'm Sanskriti Rai. I turn business problems into AI and data systems that run in production, and I lead the engineers who ship them. I went from Business Analyst to Cloud & Data Architect in 7 months.

0K+messages/day: the AI support platform I led is built for this

0specialist AI agents across 3 LLM providers

−0%COD order cancellations with an AI voice IVR

−0%reporting turnaround with a GCP data pipeline

0 mofrom Business Analyst to Architect

Most people bring one. I bring the overlap.

Business analysis, cloud data engineering and applied AI usually sit with three different people. My edge is that they meet in one person, so nothing gets lost between them. Tap a region to see what lives there.

◎And around all of it, leadership. I write the BRD, set priorities with leadership, lead a team of 4 engineers, and explain the impact to the people who need to act on it.

Built, shipped, and measured.

From a multi-agent AI platform I came up with and led, to a hardware-connected audio app I built alone.

Flagship · Idea & project lead

Garry — an entire support desk, rebuilt as AI.

Concept Kart's customers ask about orders, deliveries, returns and products on WhatsApp and email. I came up with Garry, wrote the business case and BRD/FRD, designed the architecture, and led a team of 4 engineers to production.

12specialist AI agents

3LLM providers working together

12languages, incl. Hinglish

20K+messages/day it's built for

0lost messages (durable queues)

3,800×faster inbox queries (14.5 s → 3.8 ms)

Grounded, not guessing: RAG over the live Shopify catalogue, plus real-time order and shipment lookups.
Gets smarter: a self-learning loop turns staff replies into approved answers.
Leadership sees it: an SLA dashboard (first response, resolution, AI containment) and Voice of Customer insights.

Next.jsTypeScriptSpring BootPostgreSQLRedisPub/SubClaudeGeminiOpenAIDocker

How a message travels

Customer writes inWhatsApp Cloud API or Gmail, in any of 12 languages

Durable ingestionPub/Sub, retries, circuit breaker and a dead-letter queue. Nothing is dropped.

Planner routes itAn LLM router picks the right specialist

Specialist agents answerGrounded in live store data and the knowledge base

orderstrackingcancelreturnswarrantyproduct searchfreeform + RAGescalation

Answer checkEnsemble vote plus an LLM-as-judge before anything is sent

Reply or hand offCustomer gets the answer, or a person takes over with full context

Solo build · Product, frontend & AI

Audiocular — sound tuned to your ears.

A web app that talks straight to DAC/IEM hardware from the browser, measures how you hear, and recommends EQ that fits you. I designed and built it end to end, on my own.

Hardware in the browser: WebHID connects DAC/IEM devices directly; the PEQ editor draws live graphs across 5 filter types.
A hearing test that means something: a clinical-style 7-frequency audiogram that runs in the browser and converts dBFS thresholds to dB HL, then compensates with EQ.
"Listeners like you": a GMM similarity model over filter curves recommends presets, alongside genre-blended EQ.

ReactViteTailwindWebHIDFastAPINumPySciPyPandas

PEQ · live previewYour hearing profile

Data platform

−70%

Reporting turnaround

Replaced scattered manual reporting with one GCP pipeline. It's now the single source of truth for operations, marketing and leadership across 200K+ orders.

Meltano→GCS→BigQuery→dbt→Dashboards

Cloud Run JobsGemini enrichment

Business × AI

−90%

COD order cancellations

I spotted the leak, gathered requirements and launched an AI voice IVR that confirms cash-on-delivery orders before they ship.

TwilioElevenLabsRequirements

Digital analytics

0→1

Tracking, built from scratch

Set up Google Tag Manager and connected GA4, Search Console and Google Ads conversion tracking. I was the company's main point of contact with the Google account manager.

GTMGA4Search ConsoleGoogle Ads

Smart India Hackathon 2024 · National finalist

SmartSpecMatch

An AI system that matches product specifications across 10,000+ procurement records with 90%+ top-5 accuracy. It reached the top 5 nationally, out of 86,000+ entries.

TF-IDFCosine similarityFuzzy matchingFastAPI

View on GitHub ↗

Big data

E-Commerce Behavior Analysis

Analysed 42M+ user events for behaviour patterns and cut storage by 60% (5.3 GB → 2.1 GB) by moving from CSV to Parquet with parallel processing.

DaskParquetPandas

View on GitHub ↗

Breadth is easy. Depth is the point.

Four disciplines I don't just know the names of. I've shipped production work in each of them.

Business Analysis

Start with the problem, not the tech.

Requirements gathering and BRD/FRD writing
Process mapping across operations, logistics and support
Gap and root-cause analysis, SOPs, workflow design
ROI cases that leadership can act on

BRD/FRDSOPsKPI design

Cloud & Data

Data people trust, on time.

ELT on GCP: Meltano ingestion, dbt models, BigQuery warehouse
Scheduled Cloud Run jobs replacing manual reporting
Event-driven ingestion with Pub/Sub and dead-letter queues
Query and index tuning at production volume

BigQuerydbtMeltanoPostgreSQLSQL

Applied AI

AI that holds up in production.

Multi-agent orchestration with planner routing
RAG with embeddings and scoped hybrid retrieval
Multi-LLM ensembles and LLM-as-judge quality checks
Classical ML: GMM similarity, TF-IDF, Random Forest

ClaudeGeminiOpenAILangGraphscikit-learn

Leadership

Teams that deliver.

Led a team of 4 engineers from idea to production
Set priorities with leadership; ran founder review cycles
Main point of contact with Google as a partner
Taught AI/ML to 100+ working professionals

DeliveryStakeholdersMentoring

Business brain. Architect's hands. AI at scale.

Most people bring one. I bring the overlap.

Built, shipped, and measured.

Garry — an entire support desk, rebuilt as AI.

How a message travels

Audiocular — sound tuned to your ears.

Reporting turnaround

COD order cancellations

Tracking, built from scratch

SmartSpecMatch

E-Commerce Behavior Analysis

Breadth is easy. Depth is the point.

Business Analysis

Cloud & Data

Applied AI

Leadership

Every step built toward this.

Cloud & Data ArchitectFull-time

Business AnalystFull-time

FinTech FacultyPart-time

Data AnalystInternship

Smart India Hackathon 2024 · National Finalist

Artificial Intelligence InternInternship

Data Analytics & Power BI TraineeInternship

B.Tech, Artificial Intelligence & Data Science

Employee of the Month

Smart India Hackathon

Professionals mentored

Certified

Let's build something that ships.