Atheendre "Ram" Ramesh

Atheendre
"Ram" Ramesh

I build real software and I want to operate, decide, and lead. My edge is technical depth in the service of business judgment.

CS undergraduate, RV University Open-source contributor People lead

I've run a function, not just contributed to one.

Head of Volunteer Management

Robin Hood Army · Bangalore

Leads volunteer operations for Robin Hood Army's Bangalore chapter — owning recruitment, onboarding, coordination, and retention of the volunteer base end-to-end. Responsible for deploying volunteers reliably across food redistribution drives and for building the systems that make a large, distributed team actually show up and deliver.

Problems I've owned end-to-end.

The problem

StripeFlow — Enterprise Payment Platform

Enterprise teams need payment infrastructure that doesn't fail at scale — most hobby implementations collapse under load, lack observability, or can't handle subscriptions alongside one-off charges. I designed and shipped a full-stack payment platform from scratch: REST API backend, real-time webhook delivery with retry logic, recurring billing, multi-currency support, and a live analytics dashboard.

→ 1,500+ transactions/sec at peak · 45 ms avg response · 99.9% uptime · 95%+ test coverage

Java Spring Boot React PostgreSQL Kubernetes

The problem

AI Loan Eligibility Engine

Matching loan applicants to eligible products required manual analyst time — unscalable as applicant volume grew. I built a fully serverless AWS pipeline that ingests bulk applicant data via S3, scrapes available loan products, runs AI-assisted matching logic across multiple workflows, and delivers personalised eligibility notifications — without a human in the loop.

→ End-to-end automation from CSV upload to personalised email — zero manual steps per applicant

Python AWS Lambda PostgreSQL n8n

The problem

Personality Classifier — Kaggle Competition

Top competitors were clustering at 96–97% accuracy; the ceiling was statistical noise, not capability. I built a systematic multi-model ensemble — XGBoost, AutoGluon, CatBoost, SVMs, and neural nets — then analysed where models diverged to find the performance edge rather than just stacking outputs blindly.

→ 97.5% accuracy — top of leaderboard in a dense competitive field

Python XGBoost AutoGluon CatBoost

The problem

RL Agent — Chrome Dino (Dueling Double DQN)

Training an agent from raw pixels is unstable — naive DQN diverges or learns brittle policies. I implemented Dueling Double DQN with Prioritised Experience Replay in PyTorch: separating value and advantage estimation, and focusing training on high-impact transitions rather than uniform sampling.

→ Agent learns a stable jump policy from visual input alone; ships with pre-trained weights

Python PyTorch OpenCV

The trajectory.

I started building software because I wanted to make things that actually work. Along the way I realised the most interesting problems aren't purely technical — they're about judgment, coordination, and deciding what to build in the first place.

That's the move I'm making deliberately. I have enough engineering depth to know when a technical decision is really a product or strategy decision, and to earn credibility with builders. Now I want to be in the rooms where the bigger calls get made — with founders, operators, and investors shaping what comes next.

I'm based in Bangalore, graduating in 2027, and genuinely energised by hard problems in fast-moving environments. If you're building something that matters, let's talk.

Atheendre"Ram" Ramesh

I've run a function, not just contributed to one.

Problems I've owned end-to-end.

StripeFlow — Enterprise Payment Platform

AI Loan Eligibility Engine

Personality Classifier — Kaggle Competition

RL Agent — Chrome Dino (Dueling Double DQN)

The trajectory.

Let's talk.

Atheendre
"Ram" Ramesh