SAMPURN RATTAN JAINAI Architect
Enterprise AI architecture, delivery & advisory

Applied
Enterprise AI
Architect

Eleven years turning research-grade AI into production systems enterprises bet their operations on. Currently Director of Data Science at GeeksforGeeks, leading a team of 20; formerly Solution Architect at Lumiq, architecting AI for India's largest banks and insurers.

$2 → $0.1Underwriting cost per case, at a top life insurer
99.1%Of platform queries resolved with zero human touch
$3.2MClosed deal value solutioned in 2024
$1.5M/moEnergy savings enabled for a Fortune 500 utility
Multi-agent systemsKnowledge graphsHybrid RAGDocument AIAgentic orchestrationSmall language modelsQLoRA fine-tuningVector search at scaleSemantic re-rankingZero-touch decisioningStraight-through processingGuardrails & evalsExplainable AIHuman-in-the-loop designIntelligent document processingRetrieval evaluation (RAGAS)LLMOpsModel observabilityStreaming feature pipelinesReal-time inferencePropensity modellingRisk & underwriting AISpeech & sentiment analyticsRecommender systemsMulti-agent systemsKnowledge graphsHybrid RAGDocument AIAgentic orchestrationSmall language modelsQLoRA fine-tuningVector search at scaleSemantic re-rankingZero-touch decisioningStraight-through processingGuardrails & evalsExplainable AIHuman-in-the-loop designIntelligent document processingRetrieval evaluation (RAGAS)LLMOpsModel observabilityStreaming feature pipelinesReal-time inferencePropensity modellingRisk & underwriting AISpeech & sentiment analyticsRecommender systems
01

AI Products

Owned end to end — architecture, team, roadmap, outcome
01

Leo — LiteCone

Multi-agent insurance underwriting & risk assessment · LiteCone platform, Lumiq

Oct 2024 – Jun 2025Led development & deployment
CrewAIQLoRA SLMsLangChainAWS BedrockQdrantKubernetes
litecone.ai

An AI coworker that underwrites life insurance end to end for one of India’s largest insurers — reading the case in its native form and returning a decision cited, line by line, to the insurer’s own underwriting manual.

  • Led development and deployment of LEO, the agent-driven underwriting platform now live in production for over a year.
  • Fine-tuned domain-specific SLMs to assess medical, financial and general insurance risk against the client’s manuals.
  • Built a modular architecture so the agent slots into legacy policy systems and extends to new roles without a rewrite.
  • Designed for audit from the start: every decision replayable, grounded to client data, with configurable autonomy.
$2 → $0.1Cost per proposal
48% → 72%Straight-through rate
100×Faster processing
02

Drishti Document AI

In-house intelligent document processing platform · Lumiq

Jan 2020 – Jan 2024Built and owned the platform
PyTorchAmazon TextractTransformersNLPDockerAWS

An enterprise-wide IDP engine that replaced armies of back-office operators. One platform for every document type an insurer or bank touches — scanned, handwritten, watermarked, sideways.

  • Built the full pipeline: quality analysis, preprocessing, orientation correction, text detection and extraction, classification, information extraction, face and signature extraction and verification.
  • Template-agnostic and self-learning — new document classes understood in a day, not a quarter, via continuous reinforcement.
  • Shipped as a real-time API deployable on cloud, hybrid or on-prem, including air-gapped environments.
  • Powered onboarding and KYC, bank account validation, motor invoice automation, health claims and Supreme-Court-mandated Aadhaar masking.
−70%Operational expense
−50%Turn-around time
2.5M+Documents processed
−97%Human effort in PII redaction
03

Article to Asset Pipeline

Multimodal learning-asset generation · GeeksforGeeks

Feb 2026 – PresentEnd-to-end roadmap owner
SLMsLlamaIndexGoogle AI SDKMLflowLangSmith

Turns any GeeksforGeeks article into the formats learners actually want — interactive, practice-ready and narrated — without a content team touching it.

  • Playground: converts article content into interactive visual walkthroughs for any concept.
  • Preparation: auto-generates summaries, interview questions, quizzes and coding problems from the source article.
  • Playback: uses small language models to generate video and audio narration at catalogue scale.
+60%GfG Premium sales since launch
+18%Article ad-sales revenue
04

Tikali — Power Savings

Real-time energy analytics · Fortune 500 utilities client

Jun 2015 – Jan 2022Data science lead
SparkTime-series MLKafkaPlotly DashAWS Redshift

An energy-consumption profiling platform ingesting up to a terabyte a day from global sites, and telling each site where its money was leaking — in near real time.

  • Ingests and analyses near-real-time telemetry published from sites worldwide, up to 1 TB per day.
  • Profiles consumption against historical patterns and ambient weather to surface alerts and exceptions as they happen.
  • Quantifies potential savings per site, turning raw meter data into an operating decision.
$1.5MSavings enabled per month
1 TB/dayTelemetry ingested
7 yrsIn production
02

Selected Engagements

03

Impact, In Numbers

11Years in data science, from IC to Director
20People in the team I lead today
$3.2MClosed deal value solutioned in 2024
99.1%Platform queries resolved autonomously
8,000+Students mentored across live cohorts
$300KAnnual search infrastructure cost removed
250%Increase in course sales after catalogue rebuild
26Enterprise programmes delivered
2.5MDocuments processed through Drishti
100×Faster underwriting processing with Leo
04

Experience

Data Scientist → Director, in eleven years
Jun 2025 – Present

Director — Data Science

GeeksforGeeks
  • Lead a 20-person data science organisation across content, courses and platform AI.
  • Built a GfG-wide hybrid RAG search system over 1M+ assets on Qdrant, replacing Google Search and saving USD 300,000 a year.
  • Consolidated a fragmented catalogue of 41 AI courses into 3 flagship programmes, driving a 250% increase in course sales.
  • Restructured and modernised the content pipeline, lifting viewership 50%.
  • Replaced a rule-based recommendation engine with an ML-driven user-based system.
  • Own the end-to-end roadmap for the marquee AI products — Y.O.G.I. and the Article to Asset Pipeline.
Jul 2022 – Jun 2025

Solution Architect — Data Science

Lumiq
  • Owned AI/ML solution design for BFSI clients end to end: requirements translation, architecture selection, deployment, monitoring and integration with legacy systems.
  • Solutioned 6 strategic opportunities contributing to $3.2M in closed deal value in 2024.
  • Led technical content for 3 RFP responses — all three advanced to the next sales stage.
  • Delivered technical demos to 20+ prospects; grew 4 key accounts into $200K of existing and new business.
  • Served as SME at 5 industry events including AWS AI Conclave 2024, generating 100+ qualified leads; authored 8 blogs and case studies.
Jun 2020 – Jul 2022

Lead Data Scientist

Lumiq
  • Built and led data science teams delivering real-time ML services and production pipelines over large-scale data.
  • Took Drishti Document AI from prototype to an enterprise-wide platform in production at multiple insurers and banks.
Jun 2017 – Jun 2020

Senior Data Scientist

Lumiq
  • Delivered risk, collections and delinquency models for lenders and NBFCs, including roll-forward prediction worth INR 3.5 crore a year in strategic savings.
Jun 2015 – Jun 2017

Data Scientist

Lumiq
  • Started on Tikali, the real-time energy analytics platform for a Fortune 500 utility — a terabyte a day of telemetry, and the beginning of eleven years in production ML.
Education

Integrated B.Tech + M.Tech, Computer Science

Jaypee Institute of Information Technology, Noida · 2010 – 2015

05

Stack

ML & Deep Learning
PythonPandasscikit-learnXGBoostPyTorchHugging Face Transformers
Gen AI & Agentic AI
LangChainLlamaIndexRAGCrewAIOllamaOpenAI SDKGoogle AI SDKQLoRA fine-tuning
Databases & Vector DBs
QdrantMilvuspgvectorChromaDBPostgreSQLMongoDB
DevOps & LLMOps
DockerKubernetesJenkinsGitHub ActionsPrometheusGrafanaMLflowDVCLangSmithAgentOpsDeepEvalRAGAS
Cloud & Infrastructure
AWS SageMakerBedrockLambdaRedshiftEKSGCP Vertex AIBigQueryAzure AI FoundryCosmos
Visualisation & Reporting
TableauPower BIPlotly Dash
06

Writing & Speaking

07 May 2026A Unified Framework for an Effective PromptGfG Article
25 Jan 2025LUMIQ Deploys Gen AI-Powered Negative News Detection and Risk Identification System for a Leading Life InsurerLumiq Case Study
13 Jan 2025From Hallucinations to Accuracy: Augmenting LLMs with Traditional ML TechniquesLumiq Blog
06 Dec 2024Demystifying RAG for Business LeadersLumiq Blog
06 Dec 2024A Practical Guide to RAG for Business LeadersLumiq Blog
25 Mar 2022Streamlining Health Insurance & ClaimsLumiq Blog
04 Feb 2022Reinventing Corporate Loan UnderwritingLumiq Blog
24 Jun 2021Automating Customer Onboarding with Lumiq’s Drishti Document AI and Amazon TextractAWS Blog
8,000+Students mentoredLive cohorts at GeeksforGeeks, across applied ML, generative AI and agentic systems.
5Industry events as SMEIncluding AWS AI Conclave 2024 — 100+ qualified leads generated from the stage.
20+Enterprise technical demosTranslating agent architectures for CIOs, CROs and chief underwriters.
07

Certifications

Certificates open as PDFs where available
AWSMachine Learning — Specialty (MLS-C01)Certified
DataCampMachine Learning Scientist in Python37 coursesAssociate Data Scientist in Python25 coursesPython Developer8 courses
DeepLearning.AIMulti AI Agent Systems with crewAIPDFPractical Multi AI Agents & Advanced Use Cases with crewAICompleted
UdemyPyTorch for Deep LearningPDFComplete Keras and TensorFlowPDFDocker MasteryPDFLinux MasteryPDF100 Days of PythonPDF

Let's build
something
measurable.

Open to senior AI leadership roles, enterprise AI architecture engagements, and advisory work. If you have a process with a defined manual and a queue behind it, there's usually a system worth building.

Send a message