DATA SCIENCE · ARCHITECTURE · APPLIED AI

Intelligence,
built for the
real world.

I’m Shivam, an ML Solution Architect building grounded GenAI, decision systems, and production ML—from the first proof of concept to dependable deployment.

Explore my work GitHub

Research-minded. Engineering-led. Impact-driven.

THE CONNECTED INTELLIGENCE SYSTEMFIG. 01

KNOWLEDGE
GraphRAGREASONING
Decision intelligenceTRUST
Explainable AIEXECUTION
Production ML

+FROM SIGNAL TO UNDERSTANDING+

12+ YEARS IN AI & ENGINEERINGResearch TO PRODUCTIONIndia WORKING GLOBALLY Currently at Code & Theory

01 / SELECTED WORK

Complex challenges.
Meaningful outcomes.

A selection of systems I’ve architected,
engineered, and brought into the real world.

ENTITY → RELATIONSHIP → UNDERSTANDING

89%MULTI-HOP FACT RECOVERY
INTERNAL BENCHMARK · 22 QUESTIONS

01 / KNOWLEDGE SYSTEMSCODE & THEORY

Connecting the dots.
Grounding the answer.

Enterprise GraphRAG with source traceability, multi-hop reasoning, and formal evaluation that separates retrieval quality from answer quality.

GraphRAGKnowledge graphsRetrieval

SIGNALS → MODELS → COMMERCIAL DECISIONS

$1.6BTRAILING-12-MONTH REVENUE
COVERED BY THE PRICING ENGINE

02 / DECISION INTELLIGENCECODE & THEORY

From market signals
to better decisions.

A production pricing engine, controlled experiments, and agentic explanations that connect model recommendations to commercial decisions.

Predictive MLExperimentationMLOps

A CORPUS OF CONTENT. A CLEARER SIGNAL.

50%+REDUCTION IN MANUAL CURATION
ACROSS A MULTI-MILLION-ARTICLE CORPUS

03 / CONTENT INTELLIGENCECODE & THEORY

More context.
Less manual curation.

Hybrid retrieval, recommendations, and LLM-assisted editorial workflows that turn a large content corpus into useful intelligence.

Hybrid searchLLM workflowsGCP

CONDITION → SEGMENT → GENERATE

SDXLMODULAR GENERATIVE SERVICES
IMAGE CONDITIONING + SEGMENTATION

04 / GENERATIVE VISIONCODE & THEORY

Creative control.
Production discipline.

Image-conditioned generation and segmentation workflows rebuilt as modular Python services with tracing, versioning, and regression checks.

Stable Diffusion XLLoRASAM-HQ

DIFFERENT LANGUAGES. ONE CONVERSATION.

0.45sAPPROXIMATE TRANSLATION LATENCY
3,000+ BILINGUAL PILOT CALLS

05 / MULTILINGUAL AIPULASTYA AI

Less lost in translation.
More human connection.

Real-time voice translation and multilingual commerce tools built to make language a bridge, wherever the conversation happens.

Speech AINLPDigital commerce

CLINICAL EVIDENCE × EXPLAINABLE LEARNING

1,200+SUBJECTS IN TOMOGRAPHY RESEARCH
KERATOCONUS PROGRESSION MODELING

06 / AI FOR HEALTHCAREAPPLIED RESEARCH

Earlier signals.
More informed care.

Deep learning on corneal tomography, developed with clinicians to study keratoconus progression and interpretable medical AI.

Computer visionClinical researchXAI

Selected outcomes from my professional work. Details and benchmark context are included in each case study.

SHIVAM DWIVEDIBASED IN INDIA ↗

02 / THE PERSON BEHIND THE SYSTEMS

A researcher’s curiosity.
A builder’s instinct.

I connect research, engineering, and the people who turn AI into something useful.

Across 12+ years, my work has taken me from healthcare research and conversational AI to co-founding Pulastya AI and architecting enterprise platforms. At Code & Theory, I lead the India ML/AI engineering team, combining hands-on architecture with mentoring, technical direction, delivery ownership, and production readiness.

My current focus is grounded GenAI: GraphRAG, agentic decision support, rigorous LLM evaluation, and controllable multimodal workflows. Alongside industry work, I’m pursuing a Ph.D. at IIT Patna in explainable multimodal AI for clinically safe decision support.

The longer story on LinkedIn

03 / HOW I THINK & BUILD

Across the stack.
Close to the problem.

Model the problem

Statistical modeling, causal inference, power analysis, controlled experiments, temporal validation, and the discipline to connect model quality to business decisions.

Python / SQL / scikit-learn / XGBoost

Design the intelligence

GraphRAG, agentic workflows, tool use, context engineering, grounded LLM evaluation, generative vision, and explainable multimodal learning.

Neo4j / LangChain / PyTorch / SDXL

Make it dependable

Prompt and model versioning, stage-level tracing, artifact validation, regression checks, and the architecture to carry AI services into dependable production operations.

GCP / AWS / Kedro / Argo / Kubernetes

04 / THE JOURNEY

Built over time.

Full experience

MAR 2025 — PRESENT

Code & Theory CURRENT

Solution Architect, Machine Learning

Lead the India ML/AI engineering team. Own architecture, mentoring, delivery, and production readiness across GraphRAG, agentic decision support, content intelligence, and multimodal AI.

JUN 2022 — FEB 2025

Pulastya AI

Co-Founder & CTO

Built voice AI, multilingual commerce, and decision systems across financial services, healthcare, and enterprise workflows.

AUG 2021 — MAY 2022

Concentrix Innovation Labs

Senior Machine Learning Engineer

Multilingual conversational AI and computer vision, integrated with real-time customer workflows.

JAN 2019 — AUG 2021

Techsophy

Machine Learning Engineer

Clinical AI research with L. V. Prasad Eye Institute, intent classification, and hybrid recommendation systems.

2014 — 2019

The foundations

AI consulting and software engineering at SDLC Corp; programmer analyst at Cognizant.

05 / RESEARCH & RECOGNITION

Curiosity, published.

Exploring how AI can become
more interpretable, useful, and trusted.

2023

AMRIT · SPRINGER

Early Detection of Keratoconus Progression with Deep Learning on Tomography Images

2019

IEEE TENCON

Identifying Correlation Between Frontoparietal Networks and Other Cognitive Networks in Autism

ONGOING EXPLORATION

Ph.D. · Explainable
Multimodal AI

IIT Patna · Expected completion 2028
Clinically safe decision support
Advisor: Dr. Chandranath Adak

ACADEMIC FOUNDATION

M.Tech · Artificial Intelligence

University of Hyderabad & IIIT Hyderabad (Research)
2017–2019

CONTINUED LEARNING

GenAI tooling & agents

MongoDB GenAI learning badges; Anthropic courses in MCP, Agent Skills, Claude Code, and the Claude API; Hugging Face MCP fundamentals.

Winner, ONDC-Bhashini Sprint 2024

Ministry of Electronics and Information Technology, Government of India

Also: Bhashini Velocity 2024 runner-up; represented the startup ecosystem at London Tech Week and GITEX.

06 / NOTES FROM THE FIELD

Thinking out loud.

All writing on Medium

A/B TESTING / STATISTICS

Have a complex problem?
Let’s make it possible.

shivampandit@icloud.com

AI architecture, applied research, collaboration,
or just an interesting idea. My inbox is open.

LinkedIn ↗GitHub ↗Medium ↗

Moments along the way

Pulastya team at the Startup Mahakumbh exhibition booth — Startup Mahakumbh

Group photo at London Tech Week’s Centre Stage — London Tech Week

The entrance to the Indian Institute of Technology Patna — IIT Patna

Intelligence,
built for the
real world.

Complex challenges.
Meaningful outcomes.