Currently in United StatesOpen to relocation
Prem Babu Kanaparthi
AI/ML Engineer building reliable, scalable AI systems across LLMs, inference, agents, and applied ML.
You got here somehow: a résumé, a DM, a 2 a.m. rabbit hole. Either way, you've landed on a live loss landscape. Scroll down, and watch the descent converge.
Currently building: Kerna for trustworthy agent execution, Cryo for pre-AI web search, and Nori and Sushi as live data products.
Selected systems I've shipped across agents, search, inference, and applied ML.
Where I've shipped.
Jan 2026 – Apr 2026
Rochester, NY
Graduate Researcher @ Rochester Institute of Technology
Official RIT appointment researching predictive world models and emergent emotional appraisal in generative agents.
- Extended Stanford's Generative Agents with a 72-feature LSTM predictive model, controlled scenarios, ablations, and statistical comparisons across 205,940 agent-step records.
- Validated statistically significant emergent emotion patterns against Gallup 2024 human baselines and documented model behavior, limitations, and failure modes in a research poster.
Feb 2024 – Jul 2024
Newark, CA
Generative AI Engineer @ Concentrix + Webhelp
Owned production LLM inference routing and evaluation for customer-support automation.
- Owned the production LLM inference routing layer for 50K daily requests across 3 foundation models on AWS Bedrock and SageMaker; used LiteLLM, provider fallback, and latency-aware routing to cut p95 latency from 4s to 1.5s.
- Built production LLM evaluation and safety gates with CloudWatch logging, hallucination checks, drift monitoring, and a 500-case test set; reduced incidents by 42% and MTTD by 35%.
- Reduced monthly inference spend from $45K to $37K through cost-aware routing, prompt caching, and provider fallback while maintaining 95%+ task success.
Aug 2023 – Jan 2024
Bengaluru, India
Data Science Intern @ AlphaBits Technologies
Rebuilt a Python search-ranking experimentation and offline evaluation workflow.
- Rebuilt preprocessing, feature generation, SQL-backed analysis, and evaluation across 5 model variants; cut iteration time by 90% and improved relevance by 10%.
- Created an offline ranking evaluation harness tracking relevance, error cases, and preprocessing differences across model variants before deployment.
Aug 2022 – Aug 2023
Bengaluru, India
ML Engineer Intern @ iNeuron AI
Built learner dropout-risk and phishing URL classifiers with feature and data-quality pipelines.
- Developed XGBoost classifiers for learner dropout risk (0.86 AUC across 12,000 learners) and phishing URL detection (20+ URL/domain features, 92% accuracy).
- Built a behavioral feature pipeline converting raw activity logs into 24 rolling-window features with drift and quality checks for retraining.
May 2021 – Jun 2022
Bengaluru, India
Software Developer Intern @ Exposys Data Labs
Built and tested full-stack features for an internal inventory and order-management application.
- Built reusable React and Bootstrap interfaces and Node.js/Express REST APIs backed by PostgreSQL; client-side validation reduced order-entry errors by approximately 25%.
- Automated weekly SQL reporting and assisted API refactoring and database indexing that improved retrieval speed by approximately 20%; added Jest/Mocha tests with over 70% coverage on assigned modules.
⚠ // a local minimum
For a while I optimized for the safe gradient: the projects that were comfortable, the metrics that were easy to move. They worked. They just weren't the global minimum.
The way out was never a bigger step in the same valley. It was a change of landscape: harder problems, real production constraints, research I couldn't fake. Momentum, it turns out, is what carries you out of a place that's only locally good.
Published research.
2024
Preprint
Lightweight Channel Attention for Efficient CNNs
Prem Babu Kanaparthi
Designed and evaluated a lightweight channel attention module (LCA) achieving competitive accuracy with negligible parameter and latency overhead on ResNet-18 and MobileNetV2.
You've reached the bottom of the descent.
If you scrolled this far, you're basically done running inference on me. So, what are you trying to build? I'm open to AI and ML Engineer roles, research collaborations, and the occasional weird side project. Fastest path to me is email.