CMU Heinz  ·  MISM  ·  Class of 2026

Rakshanda Hassan Abhinandan

AI Engineer, ML Researcher & Data Engineer

Building and researching AI systems, backed by 6+ years of production data engineering.

Master's student at Carnegie Mellon University, specialising in Business Intelligence and Data Analytics. I am currently a research assistant to Prof. Deva Ramanan and Prof. John Galeotti, where I am first author on a paper accepted to ECCV on how vision language models read images and text together. This past summer I also worked as an AI Engineer Intern at Foresight RCM, building AI systems for healthcare claims. Previously a Data Engineer at Oracle Financial Software Services for 6+ years, building fault tolerant ETL pipelines, automation frameworks, and ML systems across 20+ enterprise modules globally.

View Research Get in Touch
1
First Author ECCV Paper
6+
Years Experience
80%
Cycle Reduction
20+
Enterprise Modules
Rakshanda Hassan Abhinandan at CMU
🎓 CMU Heinz
🔬 AI Researcher
🧠 AI Engineer
🏛️ Oracle Alum
01

About Me

I'm currently a Master's student at Carnegie Mellon University – Heinz College, pursuing a Masters in Information Systems Management with a specialisation in Business Intelligence and Data Analytics. My coursework spans deep learning, unstructured data analytics, distributed systems, and advanced databases.

I am currently a research assistant to Prof. Deva Ramanan and Prof. John Galeotti at CMU, studying how vision language models actually read images and text, and I am first author on a paper accepted to ECCV that comes out of that work. This past summer I also worked as an AI Engineer Intern at Foresight RCM, where I built AI systems that help healthcare teams catch claim denials before they happen.

Before CMU, I spent 6+ years as a Data Engineer at Oracle Financial Software Services in Bengaluru, building Python automation frameworks that cut deployment cycles by 80%, engineering fault-tolerant ETL pipelines on OCI, and creating real-time monitoring dashboards serving 20+ enterprise modules.

I'm passionate about the intersection of data engineering and machine learning — from pipelines that make data reliable to models that make data speak. Outside of work, I've led 300-person townhalls, published IoT research, and co-founded a DevOps club that mentored 80+ students.

6+
Years professional
experience
80%
Deployment cycle
reduction
60%
ETL throughput
improvement
20+
Enterprise modules
automated
02

Work Experience

Carnegie Mellon University
Research Assistant, Vision Language Models
May 2026 – Present
Pittsburgh, USA
  • Research assistant to Prof. Deva Ramanan and Prof. John Galeotti, studying how vision language models process image and text tokens together.
  • First author on Ask Twice, Look Twice, accepted to ECCV, which diagnoses the question first paradox in VLMs using logit lens and attention probe analysis.
  • Proposed question echoing, a training free fix that restates the question on both sides of the image, lifting Winoground group accuracy by up to 19 points with no architecture changes or fine tuning.
  • Ran large scale interpretability experiments across multiple open VLMs to trace how attention shifts when the question is echoed after the image.
Foresight RCM
AI Engineer Intern
May 2026 – Aug 2026
Remote
  • Built an AI powered at risk accounts receivable watchdog that flags claims nearing payer filing and appeal deadlines, tied directly to a dollars saved metric.
  • Built a modifier suggestion model that reads NCCI edits and procedure combinations to auto suggest the correct claim modifiers before submission.
  • Governed the pipeline that turns unstructured payer policy documents into a compiled payer policy graph, and codified the base tree contract, source hierarchy, and payer overlay boundaries that the graph is built on.
  • Designed an auditable feedback loop that attributes downstream claim outcomes, including 277CA, TA1, and 999 acknowledgments, clearinghouse and payer rejections, 835 or ERA adjudication, CARC or RARC denials, and appeal results, back to the original claim decision, then turns repeated patterns into shadow rule proposals gated behind human review before anything goes live.
  • Wired payer provider directory evidence into the pre encounter eligibility check UI so reviewers see coverage evidence before a claim is submitted.
  • Productionized and hardened Foresee, the company's internal AI chatbot, adding multi turn context, PHI safe data access, and chart visualizations to a role scoped, read only assistant used across the org.
  • Shipped Phase 1 of a claim code verification feature for Foresee that validates CPT, HCPCS, ICD 10, modifier, and NDC codes and checks ICD 10 to CPT pairing compatibility, and designed Phase 2 to extend this into a full medical necessity verdict across roughly 55 clinical policy trees.
Oracle Financial Software Services
Data Engineer — Quality Analyst L2 (ProdDev)
May 2020 – Dec 2024
Bengaluru, India
  • Built automation frameworks reducing deployment cycles by 80%, accelerating rollouts across 20+ enterprise modules using Python, SQL, and PL/SQL.
  • Engineered scalable JSON, SQL, and flat-file validators, shrinking verification time from 4–6 hours to under 20 minutes while significantly increasing schema anomaly detection.
  • Initiated fault-tolerant ETL pipelines on OCI — 60% ingestion throughput gain, 40% reduction in pipeline failures, automated schema drift monitoring with historical pattern logging.
  • Integrated REST APIs for cross-system data synchronisation, achieving 50% improvement in data accuracy and eliminating manual reconciliation across eight teams.
  • Designed Jenkins CI/CD pipelines for versioning, automated testing, and deployments, increasing release reliability by 35%.
  • Built Python monitoring dashboards with predictive alerting and exception clustering; reduced post-deployment defects by 22%.
  • Led centralised Bug Database analytics, reducing customer-facing issues by 25% through anomaly detection and root-cause grouping.
Altran Technologies — Client: Vodafone UK
Software Engineer
Oct 2018 – May 2020
Bengaluru, India
  • Built and optimised a network inventory management solution, improving Vodafone's 3G/4G asset accuracy by 30% and strengthening compliance and audit processes.
  • Established PL/SQL and Java-based metadata automation workflows, cutting redundant manual steps by 40% and improving operational throughput across daily loads.
  • Supported transmission planning for nationwide 3G/4G rollouts serving 5M+ subscribers, enabling accurate link forecasting using KPI and historical metrics.
  • Coordinated across network, data, and ops teams, reducing cross-team handling errors by 18%.
03

Research & Publications

First Author · Accepted to ECCV

Ask Twice, Look Twice: Prompt Echoing Resolves the Question-First Paradox in Vision-Language Models

Vision language models answer better when the question comes after the image than before it, even though putting the question first should help the model know what to look for. Working with Prof. Deva Ramanan and Prof. John Galeotti, I used logit lens and attention probe analysis to show why: an early question does steer perception, but the answer token stops attending back to it once hundreds of image tokens sit in between. The fix, question echoing, restates the question on both sides of the image so one copy steers perception and the other is read out at answer time. It needs no architecture changes and no fine tuning, and it lifts Winoground group accuracy by up to 19 points.

Vision Language ModelsInterpretabilityLogit LensAttention ProbesTraining Free
Read the paper on arXiv →
IJCSMC Vol.7 Issue 3 · 2018

Wireless Sensors Network based Metering and Alerting System for Residential Resource Consumption

Published during my undergraduate studies. Spearheaded an IoT enabled energy metering system built on wireless sensor networks, with real time analytics, anomaly alerts, and consumption forecasting that delivered a verified 15% reduction in residential utility wastage.

IoTWireless Sensor NetworksAnomaly DetectionForecasting
Read the paper (PDF) →
04

Featured Projects

Deep Learning · Speech

Phoneme State Classification (MLP)

Built a multilayer perceptron to classify phoneme states from Mel spectrogram features for speech recognition. Applied systematic hyperparameter tuning across context window sizes and architecture depth to improve generalisation on unseen speech data.

PyTorchMLPMel SpectrogramsHyperparameter Tuning
View Code →
Deep Learning · Computer Vision

Face Recognition & Verification (CNN)

Built a ResNet-50 from fundamental PyTorch operations for open-set face recognition, classifying 8,631 identities, then using the learned embeddings to verify pairs of faces never seen during training. Reached a 3.14% Equal Error Rate and 97.2% verification accuracy. Methodology extends to surveillance, retail product recognition, and vehicle identification.

PyTorchResNet-50Open-Set RecognitionFace Verification
View Code →
AI · Healthcare

Medicare – Clinical Decision Support

Developed an AI-based risk prediction tool to identify high-risk maternity and diabetic patients using clinical parameters. Implemented automated patient prioritisation and a real-time simulation module, improving triage efficiency by 30%.

PythonPredictive ModellingRisk StratificationReal-time Simulation
View Code →
ML · Real Estate

Real Estate Property Price Predictor

City-wide property price forecasting using Random Forest regression. Integrated a Q&A bot powered by TinyLlama LLM for natural-language property search, letting users query listings and price estimates conversationally.

Random ForestTinyLlama LLMNLPPython
View Code →
Open Source · LLM

Prompt Driven Development (PDD)

Open-source contributor — enhanced code generation workflows using Gemini APIs, adding multi-file prompt execution, test scenario coverage generation, and improved usability for prompt-driven engineering tasks.

Gemini APIPrompt EngineeringPythonOpen Source
View Code →
IoT · Research

WSN Energy Metering System

Spearheaded an IoT-enabled energy metering system via wireless sensor networks — real-time analytics, anomaly alerts, and consumption forecasting. Achieved a verified 15% reduction in residential utility wastage.

IoTWSNAnomaly DetectionForecasting
Data · Streamlit

CMU Transportation Comparison Tool

A Streamlit app for CMU's campus community to compare shuttles, public transit, and ride-sharing in real time. Features a multi-modal route planner, live transport tracking, and integration of multiple external data sources.

PythonStreamlitReal-time APIsRoute Planning
View Code →
Deep Learning · Generative AI

Denoising Diffusion Probabilistic Model (DDPM)

A team project for CMU's Intro to Deep Learning course. Implemented a U-Net based DDPM from scratch for image generation, including the full training pipeline and sampling from noise, trained on a subset of ImageNet100.

PyTorchDiffusion ModelsU-NetGenerative AI
View Code →
05

Skills & Technologies

Programming
PythonJavaSQLPL/SQL
AI Engineering & LLMs
Vision Language ModelsMechanistic InterpretabilityLogit LensAttention ProbesRAGLLM GuardrailsHugging FaceAI Chatbots
ML & Deep Learning
MLPCNNsPyTorchTensorFlowScikit-learnPrompt Engineering
Data Engineering
ETLData ModelingAPI IntegrationData WarehousingSchema Drift Monitoring
Databases & Cloud
OracleMySQLNoSQLADB / ADWOCIAWS
DevOps & Tools
KubernetesJenkinsGitCI/CDJIRAPostmanAnaconda
Analytics & Viz
PandasNumPyStreamlitPredictive AnalyticsData Visualization
06

Education

Master of Information Systems Management
Carnegie Mellon University — Heinz College
Aug 2025 – Dec 2026 · Pittsburgh, USA · Business Intelligence and Data Analytics
CourseworkIntro to Deep Learning · Machine Learning · Data Visualization · Unstructured Data Analytics · Statistical Programming · Advanced Business Analytics · Advanced Databases · Distributed Systems · Digital Transformation
B.E. in Computer Science
Malnad College of Engineering (VTU)
Aug 2014 – Jul 2018 · Hassan, India
Foundation in computer science fundamentals, data structures, algorithms, and systems programming. Published IoT research in IJCSMC during undergraduate study. Co-founded the DevOps Club that mentored 80+ students in development, version control, and interview preparation.
07

Certifications

☁️
OCI Certified Foundations Associate
Oracle · 2021 & 2023
🐍
Intro to Data Science in Python
University of Michigan
📊
Python Data Analysis & Visualization
Rice University
🗄️
Oracle Advanced PL/SQL Certified Professional
Oracle
💾
Complete SQL Bootcamp
Udemy · 2020
08

Achievements & Leadership

🎤
Oracle Townhall Host

Hosted Oracle Townhall 2020 with 300+ attendees; led cultural, organisational & engagement initiatives.

👩‍💻
DevOps Club Founder

Founding member of DevOps Club (MCE); trained 80+ students in development, version control, and interview preparation.

Hello
Get in touch

Let's Connect

I'm currently open to internship and full-time opportunities in data engineering, ML engineering, and data science. Whether you have a question or just want to say hello — my inbox is always open.

Say Hello LinkedIn GitHub +1-412-214-2560