// Full-stack engineer · ISS STOXX · Mumbai

At ISS STOXX since 2024

Omkar
Mahajan

Over the past 2 years, I've built backends, data pipelines and MCP tooling that hold up at enterprise scale.

PythonServerless FrameworkAWSGraphQLFastAPIMCP · RAGTypeScript
Download résuméHire me

$ cat focus.txt

01

Experience

Current

ISS STOXX

Software Developer (Analyst)Jul 2024 — Present · Mumbai

LLM infrastructure

MCP · pgvector · RAG · GraphQL

  • I built the MCP gateway that lets 500+ enterprise clients query our GraphQL backends in natural language from inside their own LLM workflows. It answers 10,000+ queries weekly.
  • Getting an LLM to write correct queries against a 1,000+ type GraphQL schema takes more than a prompt, so I added RAG schema discovery on pgvector and a deterministic validator. Invalid queries dropped by 50%.

Product at scale

React Native · Redis · Python · GraphQL

  • Shipped MarketPro, our React Native client app, with an eight-person team — 20,000+ institutional users onboarded in under six months serving them on Web and IOS platforms.
  • I keep the GraphQL and Python APIs underneath it fast and boring: 25,000+ people a day, 99.9% uptime, no surprise data drift between server and client.

Web Development

Meilisearch · AgGrid · Highcharts

  • Search took 2.5 seconds. We moved it onto Meilisearch and it now answers in 0.25 seconds — fast enough to search through 10M+ financial records as you type.
  • Used AgGrid and Highcharts to build a data visualization dashboard that lets clients explore their data in real time, with interactive charts and customizable tables.

Data platform

Docker · Glue · Athena · S3

  • Wrote a Dockerized migration service to move 10M+ historical records from MySQL Aurora to Athena — weekly view generation got 40% faster as a result.
  • We own the ETL that turns 100GB+ of weekly data into an S3 data lake, which we and other teams consume as a GraphQL API rather than raw tables.
  • Created sync workflow that allows user-specific data to be in sync with our data lake, so that we can provide a unified view of the data to our clients.

Productivity

MCPs · Agents · Graphify

  • Standardized AI-assisted development workflow, that automates repetitive tasks for our team, closing tickets 40% faster — roughly giving 16 hours back each for each developer weekly.
02

Tech stack

Languages

PythonJavaScriptTypeScriptSQLJava

AI & LLM tooling

RAGMCPGemini APIClaudeCopilot AgentsPrompt engineering

Cloud & data

AWS AuroraAthenaGlueS3DynamoDBPostgreSQLMySQLMeilisearchRedis

Frameworks & tools

Next.jsReact NativeFastAPIDjangoGraphQLReduxDockerLinuxGit
03

Projects

01AI coding platform

LeetClone

An AI-powered coding platform: sub-second execution across 5+ languages via CodeBox, with an adaptive hint system that scaffolds logic incrementally instead of handing over the answer.

  • RAG over 500+ indexed problem contexts and solution patterns sharpens AI debugging precision.
  • Supports multiple programming languages including Python, JavaScript, and TypeScript.
  • Incremental hinting raised problem resolution rates an estimated 30%.
Next.jsPythonFastAPICodeBoxGemini APIRAG
02Medical imaging

Bone Fracture Detection

A peer-reviewed deep learning approach to cervical spine fracture detection, published and deployed as a working diagnostic tool for practitioners.

  • CNN trained on 800K+ radiological scans — 95% detection accuracy, 0.92 F1.
  • Django app lets practitioners run the model on CT scans, cutting preliminary screening time 60%.
  • Authored and published the accompanying research paper.
PythonTensorFlowCNNDjangoMedical imaging
03Mobile

MASA

Developed a fine-tuned YOLOv8 (You Only Look Once) model for fish detection and counting onboard deck of fishing vessels

  • Performs real-time fish detection and counting on the deck of fishing vessels.
  • Achieves high detection accuracy even in challenging lighting and weather conditions.
  • Mobile app built for amateur fishermen for identifying potential fishing zones and weather conditions.
React NativeExpoFirebase
04

Papers published

2023

International Conference on Futuristic Technologies (INCOFT) (IEEE Xplore)

Feasibility of Cervical Spine Fracture Detection Using Non-Conventional Slices

Cervical spine fractures are a time-critical emergency where a delayed diagnosis risks paralysis, so a CAD system that flags fractures faster helps radiologists cut diagnostic and human error. Proposes a CNN approach that runs a binary fracture / non-fracture classification on sagittal and coronal slices reconstructed from axial CT scans. Trained as a proof of concept on the RSNA dataset — 2,018 records, near-balanced positive and negative cases.

Peer reviewedMedical imaging
Paper ↗

2024

Journal of Electrical Systems

YOLOv8 based fish detection and classification on fishnet dataset

Makes the case for an electronic monitoring system built for the marine fishing sector, where outdated data collection and the absence of guiding applications leave real gaps. Surveys existing fishing apps to pin down those gaps, then proposes a YOLOv8 object detection and tracking pipeline as the response. Trains and evaluates the model on the Fishnet dataset, focused on the monitoring-generated data that would drive it in practice.

Peer reviewedComputer vision
Paper ↗
05

Education

2021 — 2024

Mumbai

B.E. Computer Engineering

Vivekanand Education Society's Institute of Technology — Honours AI & ML

Honours AI & ML

9.11 / 10

CGPA