AS/NApplied AI / 26 Available for meaningful work
AI systems / research / productKerala, India / working worldwide

Adithya S Nair / AI engineer

Build AIthat earnsconfidence.

I turn difficult AI ideas into measurable systems. My work spans medical vision, agent memory, evaluation, and products people can trust.

See selected work Read resume
HUMAN / SIGNAL
Professional portrait of Adithya S Nair, AI engineer and researcher
FrameBuildProve
Explore
question assumptionsprototype the hard partmeasure what matters
Operating thesis

A model can impress. A system must hold up.

012 yrs

shipping production AI

023

selected systems with real stakes

031

peer-reviewed research paper

04200+

people led in tech community

Interactive field lab / 01

Test the system.

Choose a layer. See what I optimise before an AI product reaches real users.

Active layer / M-01

Keep only useful state.

Memory should preserve decisions, evidence, and unresolved risk. More context is not the same as better context.

Primary measure: Retrieval precision
Selected systems / 03

Systems with stakes.

Healthcare, education, and product delivery. Three places where reliability matters more than a polished demo.

01 / 03Production medical AISelected case

Doctreen

Clinical signal, engineered for real use.

Built and maintained the AI core of a medical reporting platform across providers, clinical workflows, and the less glamorous realities of production.

Visit the work
The tension

Reliable medical AI is an evaluation problem as much as a modelling problem: outputs need guardrails, measurable quality, and predictable operating costs.

The response

Designed multi-provider LLM and RAG pipelines, DICOM vision workflows, adversarial-input guardrails, parallel benchmarks, CI/CD, and internal data tooling.

What changed

A production system that could be tested against radiologist ground truth, monitored for regressions, and improved without treating every model change as guesswork.

PythonFastAPIRAGLLM evaluationDockerCI/CD
02 / 03Education infrastructureSelected case

Vidyapath

One school system. Four very different realities.

A multi-school LMS for CBSE Classes 10 and 12, shaped around the daily workflows of students, teachers, parents, and administrators.

Visit the work
The tension

School software often becomes a collection of disconnected forms. The real challenge is making roles, data, permissions, and AI assistance feel like one dependable system.

The response

Architected role-based authentication, multi-school provisioning, assignments, grading, attendance, timetables, audit trails, and a multi-model AI routing layer.

What changed

A single platform that reduces teacher busywork while keeping generated learning material structured, reviewable, and grounded in the school context.

Next.jsSupabaseRAGGroq + GeminiZustandFramer Motion
03 / 03Founder-led product studioSelected case

CodrantLabs

Software shipped, not strategy left in slides.

A product and AI engineering studio for teams that need senior thinking, fixed scope, and a direct path from an ambiguous problem to working software.

Visit the work
The tension

Small teams do not need more ceremony. They need someone who can connect product intent, engineering choices, AI capabilities, and the realities of delivery.

The response

Co-founded a practice spanning full-stack platforms, AI agents, RAG systems, search visibility, and technical product strategy for global clients.

What changed

A deliberately small studio model: clear ownership, fewer hand-offs, and work that is designed and engineered as one continuous decision process.

Product strategyAI agentsFull-stackRAGAEODelivery
Published researchPCCDA 2026 · Springer
SPARE

Smaller model.
Serious signal.

Single-view Parameter-efficient Adapter for Radiology Reporting

A vision-language framework connecting RAD-DINO to BioGPT through a custom semantic alignment adapter, built for single frontal chest X-rays and constrained hardware.

98%reduction in trainable parameters with LoRA
79.0%BERTScore F1 on MIMIC-CXR
0.8%of the training corpus used
93.6%of state-of-the-art performance retained
Method / background

How I get to useful.

I connect product thinking, research, and engineering. I ask the difficult question early, prove the hard part quickly, and stay accountable after launch.

Frame the system

Translate an ambiguous product problem into boundaries, risks, data flows, and a buildable first version.

Architecture · Product thinking · Evaluation design
Experience

Work in the real world.

Jun 2025 — May 2026

AI Engineer / Doctreen

Montpellier · Remote

Owned production AI features, medical vision evaluation, internal tooling, and the infrastructure needed to make model changes observable.

Jan 2025 — Jun 2025

AI Research Intern / Doctreen

Montpellier · Remote

Helped establish the medical reporting architecture, model benchmarks, and security guardrails during the platform’s foundational stage.

Education

Built on research.

01
2026 — Present

MTech · Computer Science

AI & Machine LearningAmrita Vishwa Vidyapeetham

Researching agentic AI, memory, multi-agent systems, and LLM architecture.

02
2021 — 2025

BTech · Computer Science

AI & Machine Learning · CGPA 8.48Amrita Vishwa Vidyapeetham

Machine learning, deep learning, NLP, computer vision, systems, and the SPARE research work.

200+people led
Leadership / community

ACM Student Chapter, Amritapuri

Member → AI Club Mentor → Chairperson → Advisory Council

Led a 200+ member technical community for 1.5 years, organised national-level hackathons and workshops, mentored students in machine learning, and now advise the next leadership team.

Selected experiments / 05

Built to learn.

202601

Mitti Mitra

Field-ready soil diagnostics that turn vision and multi-source signals into useful crop guidance.

Computer visionFlutterFirebase
202602

LensAI

Explain any selected region of a webpage without breaking focus.

Browser APIsLLMsJavaScript
202503

Namude Yatra

A multi-agent travel planner with maps and conversational replanning.

LangChainPydeckStreamlit
202404

OptiHire

Tailored job outreach, resume gap analysis, and application tracking.

NLPLLMsData
202305

SWC from MRI

Graph representations of brain vascular flow generated from MRI data.

Medical imagingPythonGraphs
Next06

Your hard problem

The list grows wherever a difficult idea needs a working, measurable system.

Start one
Editorial illustration of an AI memory system moving through evaluation checkpointsField note / 001
New from the journal

Memory is not a feature.

Why capable agents need selective memory, observable retrieval, and evaluation across the full task journey.

12 min readAugust 3, 2026
Read the field note Browse all writing
Common questions / 06

Direct answers.

The short version — for people in a hurry and the engines answering for them.

Who is Adithya S Nair?

Adithya S Nair is an AI engineer and researcher from Kottayam, Kerala, India. He builds production LLM, RAG, agentic, and medical imaging systems, and published the SPARE radiology reporting research at PCCDA 2026 (Springer).

What does Adithya specialise in as an AI engineer?

He specialises in LLM integration, retrieval-augmented generation, agent memory, AI evaluation, and DICOM medical vision. He ships full-stack AI products with Python, FastAPI, PyTorch, and Next.js — from first prototype to monitored production.

Which production AI systems has Adithya built?

At Doctreen he built the AI core of a live medical reporting platform. He architected Vidyapath, a multi-school LMS with four role-specific portals, and co-founded CodrantLabs, a product and AI engineering studio.

What research has Adithya S Nair published?

SPARE — Single-view Parameter-efficient Adapter for Radiology Reporting — accepted at PCCDA 2026 (Springer). It cuts trainable parameters by 98% with LoRA and reaches 79.0% BERTScore F1 on MIMIC-CXR using 0.8% of the training corpus.

Is Adithya available for AI engineering roles?

Yes. Adithya is open to AI engineering roles, research collaborations, and focused product builds with real users. Email adithyasnair2021@gmail.com to start a conversation.

Where is Adithya based, and does he work remotely?

He is based in Kottayam, Kerala, India, and works remotely worldwide. He previously worked remotely as an AI Engineer with Doctreen in Montpellier, France, on production medical AI.

The next hard thing

Let's build the hard part.

I am open to AI engineering roles, research collaborations, and focused product builds with real users and real constraints.

Start a conversationSay hello
adithyasnair2021@gmail.comLinkedInGitHubResume