RR

finding the signal

Available for ambitious product work

I build useful intelligent systems that survive the real world.

Rahul is an applied AI/ML engineer who turns messy problems into working products — from resilient cloud operations and multimodal RAG to offline accessibility tools.

Rank 83 / 4,000+TECHgium national finalistMaintainer-merged PRs
Rahul / field modeMove gentlyRahul Rathnavel presenting his work

Evidence, not noise

The work has receipts.

Awards matter. Working prototypes, merged PRs, and repeatable technical decisions matter more.

83

Amazon ML Challenge

Cyber Titans reached Rank 83 with a 45.35 SMAPE final submission after repeated model and evaluation iterations.

34

TECHgium finalist teams

SmartOps reached the national final stage after a nine-month innovation journey.

816

Problems solved

Knight level. 2,570 submissions across 351 active days.

Selected systems

Built where the constraints are real.

Each story starts with the problem, keeps the hard parts visible, and avoids claims the evidence cannot support.

Brothers of Creations at the TECHgium SmartOps finals
01 / AIOps prototype

SmartOps: turn an incident into an answer.

Designed a human-reviewed AI operations loop that observes failures, diagnoses the likely cause, proposes a fix, and learns from the SRE team’s feedback.

AIOpsGNNQwenEKS
Amazon ML Challenge leaderboard showing Cyber Titans at rank 83
02 / Machine learning

Rank 900 was only the start.

A price-prediction system refined through leaderboard drops, better pooling, outlier control, and a final Rank 83.

NLPDeBERTa-v3PyTorchSMAPE
ESP32-CAM smart glass prototype
03 / Edge AI accessibility

A smarter pair of glasses should protect privacy too.

An offline assistive prototype pairing wearable visual input with a Flutter app for object, text, and face assistance.

Edge AIFlutterESP32-CAMVision
04 / Live multimodal RAG

Answers should show their evidence.

Omni RAG combines dense and sparse retrieval, then returns an answer alongside the exact PDF page that supports it — because confidence without evidence is not enough.

RAGFAISSBM25Llama 3

Open source

Maintainer-reviewed, not just published.

Small, precise fixes across mature Python libraries — the kind of work where the details actually matter.

Optional detour

Prefer a guided tour?

Take the RR Signal Route: a light, keyboard-friendly interactive map of the work, timeline, and proof points. It opens separately, so this page stays exactly where you left it.

05 / Android

Voice Surrogate

On-device bilingual smart replies for non-speaking individuals. Speech in, a considered response out — without handing a conversation to the cloud.

06 / Mobile

Raki Chat

A Kotlin and Jetpack Compose chat app built around Firebase authentication, real-time messaging, MVVM, and a deliberately maintainable structure.

07 / Research prototype

Biometric Voting

A controlled voting-system prototype exploring Solidity auditability alongside biometric verification, a Next.js interface, Flask services, and Arduino hardware.

Independent work / names respectfully withheld

Quiet builds. Real constraints.

A few projects were delivered privately, so the portfolio shows the problem space and the engineering work — never a client name, dataset, or claim that should stay confidential.

01 / Trust

Job-signal verification

A workflow for classifying suspicious job listings and matching opportunities to a supplied résumé, with structured extraction from public job links.

02 / Health research

Breast-cancer decision-support prototype

A research prototype combining tabular classification and image-led signals with nearby-care discovery. It is not a diagnostic product or medical advice.

03 / Field telemetry

Farm conditions, on a phone

A mobile view of wirelessly collected Arduino sensor readings so growers can check humidity and related field conditions without standing beside the hardware.

04 / Privacy research

Encrypted smart-grid prediction

An ML research workflow exploring predictions over protected smart-grid data, designed around the requirement that sensitive data should not be exposed just to make a model useful.

Contact / no forms of ceremony required

Let’s make something useful.

A product idea, an AI system that needs a sanity check, or a film recommendation — Rahul reads every thoughtful note.