Open to SDE and AI Engineer roles · Jan 2027

I build AI that ships, on backends that don't break.

I'm Neha, a software and ML engineer with 6+ years of production experience across India, Oman and the US. Now finishing my MS in AI/ML at RIT and building RAG and multimodal systems you can actually measure.

See what I've built → Get in touch
GitHub LinkedIn New York · US work authorized · Open to relocation
Neha Rani
Neha RaniSoftware & AI Engineer · New YorkOpen to relocation
Built production systems at
OoredooE2OpenAT&TAccentureTech Mahindra
Why work with me

Production engineer first. AI engineer now.

Most ML engineers haven't run systems for millions of users. Most backend engineers haven't fine-tuned a model. I've done both, and I bring that to every project.

6+ years

Shipping production systems at scale

Java and Python backends for telecom super-apps, global supply chain SaaS and subscriber platforms. I know what breaks at 3am and how to stop it.

OoredooE2OpenAT&TAccentureTech Mahindra
3.81

Master's in Computer Science, specialization in ML/AI

Rochester Institute of Technology. Coursework in machine learning, artificial intelligence, algorithms, computer vision and big data analytics. Graduating December 2026.

Applied ML

Fine-tuned DistilBERT and ResNet50, fused into one model, served in real time.

Evaluation first

I benchmark every retrieval and model choice, so decisions come from data.

Full stack

From FastAPI and Postgres to React dashboards and AWS deploys with CI/CD.

Three countries, one way of working

I've shipped in India, Oman and the US.

Different teams, time zones, clients and expectations. What stayed the same is how I work: I own the problem until it's fixed.

India

Bangalore · 2017 – 2022

Started at Accenture building for AT&T, a US client across time zones. Then moved to E2Open, a global supply chain platform used by 60,000+ companies.

US + global clients

Oman

Muscat · 2022 – 2024

Relocated to work on-site with Ooredoo as a Senior Software Engineer. Owned critical backend flows for a super-app serving 3M+ subscribers in the Middle East.

On-site with client

United States

New York · 2024 – now

Moved to the US for an MS in AI/ML at RIT. Building RAG, multimodal and federated learning systems, and ready to bring it all to a US team.

GPA 3.81US work authorized

Working across 3 countries and industries from telecom to supply chain to AI taught me to get up to speed fast, communicate clearly with people who think differently, and deliver no matter where the team sits.

I own problems end to end

When something breaks, I don't stop at the patch. I find the root cause and make sure it doesn't come back.

ProofLed root-cause analysis across 4 telecom services and cut repeat incidents by 35% in six months.
ProofDesigned rollback logic that took plan migration failures to zero and ended manual fixes for the ops team.

I lead and lift the team

As a senior engineer I led design decisions on critical flows and mentored newer engineers so the team could move faster without me.

LeadOwned the design of circuit breakers and timeouts that cut SIM activation failures by 43%.
MentorGuided junior engineers through code reviews, debugging production issues and onboarding to complex telecom systems.

I stay committed under pressure

High-volume campaigns, 3M users, production incidents. I'm the person who stays on it until it's stable.

ProofBuilt async pipelines with RabbitMQ that kept APIs stable during large-scale campaign spikes.

I adapt fast

New country, new domain, new stack. I've switched from telecom to supply chain to AI research and delivered in each.

ProofWent from enterprise Java to shipping a live multimodal ML app within my first two years in the US.
Featured work

Things I've built and shipped.

End to end projects where I owned the model, the API, the infrastructure and the interface.

● Live demoMultimodal MLFeb – Mar 2026

RescueAI: faster triage when every minute counts

First responders get flooded with reports during a disaster. RescueAI reads text reports and aerial images together and ranks incidents by urgency, live on a map.

89%text severity accuracy
86%aerial image accuracy
+22%ranking precision
<120msp99 latency

DistilBERT + ResNet50 fusion · FastAPI with REST and WebSockets · AWS, Docker, GitHub Actions · React + Leaflet · offline mesh relay for no-signal zones

Try the live demo ↗ View code
RescueAI · incident queueLive
Ranked by priority
94
Flooding, people on rooftopsText: critical · Image: severe
0.4 km
87
Collapsed building reportedText: critical · Image: debris
1.2 km
61
Road blocked by fallen treesText: moderate · Image: obstruction
2.8 km
28
Power outage, no injuriesText: low · Image: no damage
3.5 km
Sample incidents for illustration
● In progressRAG · LLMsApr 2026 – now

SEC-RAG-Bench: RAG you can actually trust

Most RAG demos never prove they work. This one tests every design choice against a 120-question benchmark built from real SEC 10-K filings, with ground-truth citations.

1610-K filings
120benchmark questions
6question types
4quality metrics

pgvector + BM25 hybrid search · cross-encoder reranking · Recall@k, citation accuracy, faithfulness · Redis semantic cache · Claude/OpenAI fallback · Grafana

SEC-RAG-Bench · pipelineRunning
01
QuestionFactual, multi-hop, numerical, negation…
FastAPI
2a
Dense searchpgvector
2b
Keyword searchBM25
03
RerankCross-encoder
Hybrid
04
Answer with citationsClaude · OpenAI fallback
LLM
05
Score itRecall@k · citations · faithfulness
Auto
More projects

Data engineering to ML security.

Snowflake Churn Feature Store

An end to end churn pipeline on Snowflake with lineage tracing, data validation, and a versioned feature store that removes duplicate features automatically.

SnowflakeXGBoostscikit-learn
0.817AUC, logistic regression
0.815AUC, XGBoost

Federated Learning Security

A defense against poisoned updates in federated learning. A PID controller tracks how far each client drifts from the global model and drops the suspicious ones.

PyTorchFlowerMedMNIST
20clients, 8 attackers
73.6%+accuracy under attack
Experience

Real impact, in production.

The numbers are what changed after I worked on the system.

● Muscat, Oman · 2022 – 2024
−43%SIM activation failures

Tech Mahindra · Ooredoo

Senior Software Engineer
  • Backend for a super-app with 3M+ subscribers
  • Migration failures down to zero
  • 3× faster queries, 62% faster bulk updates
  • 35% fewer repeat incidents
● Bangalore, India · 2020 – 2022
−65%database calls at peak traffic

E2Open

Application Engineer
  • Platform connecting 60,000+ companies
  • Removed all false order alerts
  • Disruption alerts 24–48h earlier
● Bangalore, India · 2017 – 2020
AT&Tsubscriber management platform

Accenture

Software Analyst
  • Full-stack features in Java and React
  • Real-time device trade-in pricing
  • Legacy Struts to Spring MVC migration
Toolkit

What I work with.

AI / ML

PyTorchHugging Facescikit-learnXGBoostOpenCVRAGLLMsFine-tuning

Backend

PythonFastAPIJavaSpring BootNode.jsRabbitMQ

Data

PostgreSQLpgvectorElasticsearchRedisSnowflake

Frontend & Cloud

ReactTypeScriptAWSAzureDockerGitHub Actions
Education
MS Computer Science, AI/MLRochester Institute of Technology · 2024 – Dec 2026
BE Computer ScienceDayananda Sagar College of Engineering · 2013 – 2017
CertificationsGenerative AI with LLMs (DeepLearning.AI) · Azure AZ-900
Let's talk

Looking for someone who can build and ship?

I'm open to full-time SDE and AI Engineer roles starting January 2027, in New York or anywhere in the US. I'm open to relocation. Email is the fastest way to reach me.