Harshitha Kolukuluru
I build high-throughput distributed systems and data pipelines, and design ML-driven workflows focused on evaluation, scalability, and reliability, making complex systems efficient and practical in real-world environments.
About Me
I'm a Computer Science graduate student at UMass Amherst working at the intersection of ML systems, applied machine learning, distributed systems, and data-intensive platforms, with a focus on building ML systems that are robust, reproducible, and effective under real-world constraints.
As a Graduate Student Researcher in the BioNLP Lab, I work on multi-agent medical simulation frameworks using LLMs, GraphRAG, and reinforcement learning. My work involves designing memory-augmented agents, reducing hallucinations in clinical interactions, and developing evaluation pipelines for intelligent systems operating in complex, dynamic environments.
At Adobe, I work on ML system orchestration for long-running workflows, focusing on improving efficiency and reliability through better routing, context management, and policy-driven execution. This experience has deepened my interest in building scalable ML infrastructure and optimizing systems under real-world constraints.
Previously, I was a Software and Site Reliability Engineer at Rakuten Mobile (Tokyo), where I built and operated large-scale distributed systems in production. I designed and deployed microservices on Kubernetes, implemented observability and infrastructure automation, and helped maintain 99.9% uptime while improving detection, deployment velocity, and overall system performance.
Across both research and industry, I enjoy bridging experimentation and production, building distributed, data-intensive systems that support applied AI and enable reliable, scalable ML workflows. I'm currently seeking opportunities in ML systems, applied AI, data science, software engineering, and site reliability engineering roles.
Experience
Graduate Research Extern
Adobe Research
Feb 2026 - Present
- Working on learned orchestration for efficient deep research agents, focusing on training policy-based controllers to dynamically prune low-value contexts in long-horizon research workflows.
- Investigating limitations of prompt-based orchestration (e.g., FlashResearch-style systems) that lead to bloated context windows, inconsistent pruning decisions, and increased inference latency.
- Designing a context-pruning and termination policy that operates over mid-research signals such as accumulated findings, context relevance, and diminishing returns.
- Exploring reward modeling strategies that combine goal satisfaction, answer quality, information gain, and latency reduction to support reinforcement learning–based post-training.
Graduate Student Researcher
BioNLP Lab, UMass Amherst
Feb 2025 - Dec 2025
- Designed and implemented a large-scale AI Hospital multi-agent medical simulation, engineering realistic patient, doctor, and nurse agents using GraphRAG-enhanced retrieval to enable dynamic, clinically grounded interactions.
- Developed advanced patient-agent behavior models and role-playing strategies, including a fact-checking pipeline to validate synthetic medical data and generate high-fidelity datasets for training and evaluating clinical LLMs.
- Built scalable multi-visit and multi-patient simulations (e.g., ChatCLIDS), enabling agents to leverage fine-grained conversation summaries, reflections, and longitudinal context for improved decision-making and cross-patient generalization.
- Prototyped a hospital-wide shared memory pool storing interaction trajectories, enabling weaker agents to learn from stronger ones via similarity-based retrieval, subgraph clustering of patient profiles, and lesson distillation using models such as GPT-4o.
Software & Site Reliability Engineer
Rakuten Mobile Inc., Tokyo, Japan
Jan 2023 – Jul 2024
- Engineered a Django-based Celery scheduling system to manage CPaaS SMS workflows, implementing concurrency controls to prevent task collisions and reducing Mean Time to Detect (MTTD) by 180%.
- Designed and built a distributed backend for collecting, storing, and processing SMS delivery receipts using Python, PostgreSQL, and Redis, improving concurrency handling and reducing data processing latency by 20%.
- Integrated backend services with a ReactJS monitoring dashboard and deployed the system via Nginx and ArgoCD, enabling real-time visibility, scalability, and high availability.
- Automated daily performance reporting to executive stakeholders by programmatically extracting metrics from Grafana, Kibana, and Elasticsearch, fully eliminating manual reporting workflows.
- Deployed and operated 6+ microservices on Kubernetes using Helm, optimized CI/CD pipelines with Argo and GitLab, and accelerated deployment cycles by 6× while sustaining 99.9% uptime.
- Built and maintained end-to-end observability pipelines using Prometheus, Grafana, and the ELK stack, delivering actionable alerting and performance insights that improved platform efficiency by 25%.
- Defined and managed cloud infrastructure using Terraform (IaC), ensuring consistent deployments and reducing configuration-related errors by 40%.
Product Management Intern
Univ.AI, Bangalore, India
Aug 2022 – Dec 2022
- Drove cross-functional collaboration across engineering, design, and web teams to align on product goals, clarify requirements, and deliver initiatives from concept to execution, fostering clear communication, accountability, and shared ownership.
- Formulated data-driven growth strategies using analytics to optimize program outreach, increasing user engagement by 30% across educational offerings.
- Authored technical content on AI trends and emerging technologies, reaching 1,000+ readers and contributing to Univ.AI's thought leadership and community growth.
Education
University of Massachusetts Amherst
Master of Science in Computer Science
CGPA: 3.967 / 4.0
Sept 2024 – May 2026
Selected coursework: Industry Mentorship Practicum (AI), Reinforcement Learning, Natural Language Processing, Advanced Machine Learning, Advanced Algorithms, Neural Networks, Applied Information Retrieval, Systems for Data Science, Statistics
Teaching Assistant: CS689: Advanced Machine Learning under Prof. Justin Domke
Indian Institute of Technology Indore
Bachelor of Technology in Electrical Engineering
CGPA: 3.7 / 4.0
2018 – 2022
Relevant coursework: Data Structures and Algorithms, Computer Programming, Computer Vision, MLOps, Digital Signal Processing, Control Systems, Digital Communication, Network Theory, Linear Algebra, Numerical Methods
Activities and Societies: DebSoc(Debating Society), Model United Nation (MUN IITI), Student International Affairs Cell (SIAC), Counselling Cell
Projects
AI Hospital / Multi-Agent Medical Simulation
Multi-agent medical simulation framework using LLMs, GraphRAG, and reinforcement learning. Designed memory-augmented agents and evaluation pipelines for intelligent agents in clinical environments.
Learned Sparse Retrieval with Vector Quantization (LSR-VQ)
Advanced information retrieval system combining learned sparse representations with vector quantization techniques to improve search efficiency and accuracy in large-scale document collections.
Two-Stage Fit-Aware Fashion Retrieval System
Intelligent fashion recommendation system that considers both style and fit preferences. Uses a two-stage approach to match users with clothing items that align with their aesthetic and sizing requirements.
Real-Time E-Commerce Analytics & Recommendation System
Real-time analytics platform for e-commerce with personalized recommendation engine. Processes streaming data to provide instant insights and dynamic product suggestions based on user behavior.
Skills
Leadership & Extracurricular
Outreach Coordinator, Co-Founder
Student International Affairs Cell (SIAC), IIT Indore
- Co-founded the Student International Affairs Cell at IIT Indore, expanding international outreach by organizing high-impact events such as the Nobel Laureates Lecture Series and collaborating with the International Affairs Office to increase awareness of global opportunities.
- Led and managed a 7-member outreach team, driving initiatives that improved student visibility and access to international programs.
Chief of Executive Board Affairs
Model United Nations (MUN), IIT Indore
- Directed the first fully virtual MUN IIT Indore with 200+ delegates, overseeing Executive Board selection and ensuring procedural integrity while maintaining event quality during the online transition.
- Coordinated outreach to 50+ schools and colleges across India as USG Public Affairs, breaking participation records with 300+ delegates, the highest in MUN IIT Indore history.
Mentor, Core Team Member
Debating Society (DebSoc), IIT Indore
- Mentored 20 students in competitive debating, strengthening argumentation, public speaking, and critical thinking skills.
- Organized and judged multiple debating competitions and events, contributing to a strong campus culture of debate and discourse.