AI Engineer: LLM Evaluation & Multimodal AI

Nikhil Reddy Yerragondu

***********@*****.*** +1-205-***-**** linkedin.com/in/yerragondu github.com/yerragondu Summary

AI Engineer specializing in LLM evaluation, multimodal AI, and AI-powered product development, with hands-on experience improving reasoning reliability, model alignment, and training data quality across production-scale AI research initiatives. Proven track record evaluating 50+ multi-step agent workflows, designing human-in-the-loop evaluation pipelines, and building full-stack AI products using Node.js, TypeScript, PostgreSQL, and modern ML frameworks. Published researcher in weakly supervised medical image segmentation and multimodal emotion recognition, combining AI research depth, product engineering, and data engineering experience to ship reliable, user-centered AI systems. Work Experience

Cheaha Infosys

Junior AI Engineer Jun 2025 – Present

• Designing and developing an AI-powered financial decision simulator aimed at augmenting human decision-making through contextual AI reasoning on real financial data, currently in active MVP development.

• Architecting full-stack application infrastructure using Node.js, TypeScript, PostgreSQL, Supabase, and Tailwind CSS, building a scalable product foundation ahead of AI model integration.

• Personalizing financial decision support by integrating a Machine Unlearning-based AI layer that adapts to individual user behavior, refines decision context over time, and enables the simulator to function as a personalized AI decision assistant, currently in active development.

• Conducted user research with 25+ real-world participants and performed competitive analysis against existing market solutions, translating findings into product requirements, feature prioritization, and MVP roadmap decisions.

• Applied a product-first, problem-first engineering approach to build a production-grade AI tool addressing real financial decision challenges beyond standard AI agent implementations.

Systems Engineer Intern (Identity & Access Management) Feb 2025 – May 2025

• Supported enterprise IAM operations across 2,000+ authentication workflows by managing user access, account lifecycle requests, and access-control processes, ensuring secure and compliant user access across systems.

• Configured and troubleshot SSO integrations using SAML 2.0, OAuth 2.0, SAML Tracer, and Postman, resolving token, assertion, and attribute-mapping issues across enterprise applications.

• Monitored authentication logs and incidents using Splunk, CrowdStrike, and ServiceNow, identifying access anomalies, authentication failures, and operational issues while supporting SLA-driven resolution. AI Engineer (Contract) Handshake Oct 2025 – Mar 2026

• Improved production-scale AI training quality across 3 research initiatives spanning LLM reasoning integrity, multimodal learning, and human- in-the-loop evaluation by designing and executing structured evaluation pipelines, strengthening model robustness, alignment, and training data reliability.

• Reduced reasoning and evaluation risks across 50+ multi-step LLM agent workflows with 25–50 steps each by systematically analyzing Plan

+ Code execution trajectories in Project Orion to detect hallucinations, logical inconsistencies, unsafe behaviors, test/dev set leakage, evaluation script manipulation, unsafe file operations, and scientifically invalid agent behaviors.

• Improved multimodal model reliability in Project Hedgehog by performing cross-modal validation across synchronized image, audio, video, and text inputs to ensure semantic alignment, temporal consistency, and robust behavior under noisy and ambiguous real-world conditions.

• Enhanced perception and grounding pipeline robustness by identifying recurring multimodal failure modes, conducting systematic error analysis, and collaborating with cross-functional research and engineering teams to improve training data quality and evaluation pipelines.

• Accelerated conversational AI training in Project Lexicon by generating structured human-human video and speech interaction datasets supporting speech, vision, discourse modeling, and human-in-the-loop evaluation. Software Developer (Founding Team) BeeP Sep 2024 – Jan 2025

• Built an early-stage secure messaging platform as part of a 2-person founding team, developing responsive frontend components and user interaction flows to improve prototype usability and product navigation.

• Enabled secure user onboarding and session management by integrating Firebase and AWS Cognito authentication workflows, supporting account creation, login, and MVP access control.

• Validated product direction by conducting user research with 50+ participants, identifying usability gaps, and translating feedback into feature improvements and UX design updates.

• Supported data-driven MVP decisions by implementing analytics tracking for user engagement and feature adoption, enabling the team to understand user behavior and prioritize product iterations.

• Accelerated early product validation by rapidly prototyping features, testing core messaging workflows, and iterating on the platform in an agile startup environment.

Data Engineering Intern (Team Lead) Talent Engines Mar 2024 – Jun 2024

• Modernized data collection for a legal recruiting firm by replacing manual scraping processes with automated Python-based extraction pipelines, enabling scalable lawyer and lateral recruiting data acquisition across 250+ websites.

• Designed and built end-to-end web scraping pipelines using Python, Selenium, BeautifulSoup, sitemaps, and automation tools to extract structured attorneys, firm, and recruiting data for downstream business workflows.

• Transformed unstructured web data into clean, structured datasets supporting Neo4j graph database relationships, n8n automation workflows, ETL processing, and recruiting intelligence pipelines.

• Led a team of 7 engineers, coordinating scraping tasks, reviewing data outputs, resolving blockers, and ensuring timely delivery of high-quality datasets aligned with business requirements.

• Communicated project progress, technical updates, and delivery status directly to leadership, bridging engineering execution with legal recruiting business goals.

Machine Learning Intern PanTech Solutions Jun 2022 – Jul 2022

• Built computer vision models using Keras and OpenCV for image and activity recognition across varied environments.

• Improved model robustness with targeted augmentation (lighting, distortion, motion blur), boosting accuracy by 20%.

• Developed visual dashboards to communicate model insights to cross-functional teams and reduced model training time by 30% through optimized workflows and experimentation.

Research Publications

SPARK: Sparse Prior Adaptive Representative Knowledge for Unsupervised Segmentation (Under Review) NeurIPS 2026

• Designed SPARK, a weakly supervised segmentation framework that addresses limited medical annotation availability by learning coherent semantic regions from sparse scribble annotations instead of dense pixel-level labels.

• Improved segmentation robustness across 7 datasets spanning 5 medical and 2 natural image benchmarks by integrating sparse semantic anchoring, Local Semantic Barycenters, and Medical Structural Consistency constraints.

• Achieved statistical parity with fully supervised baselines while reducing pancreas CT volumetric error variance by 35 and improving DSC by +0.16 over weakly supervised methods, demonstrating strong performance under sparse supervision. Multifold Fusion Attention Variant (MFAV) for Multimodal Emotion Recognition IEEE CSASE 2025

• Developed MFAV, an end-to-end multimodal emotion recognition model combining EEG, audio, and video signals to address limitations of feature-dependent and simplistic fusion-based emotion recognition systems.

• Designed attention-based intermediate fusion and dynamic modality weighting mechanisms to improve robustness across heterogeneous inputs and capture cross-modal, spatial, and temporal dependencies.

• Achieved 91.2% accuracy and 0.08 MAE, improving performance by 7.39% over baseline models while demonstrating the importance of fusion through ablation results.

Education

University of Alabama at Birmingham (UAB) - MS in Computer Science Jan 2024 – Apr 2025

• Presentations and Proceedings at Conferences: SPIE Medical Imaging 2025, IEEE CSASE 2025 Guru Ghasidas Vishwavidyalaya (GGV), India - BTech in Computer Science Aug 2019 – Jun 2023

• Presentations and Proceedings at Conferences: ICLTEM 23, and ISPEC 21 Skills

ML PyTorch, TensorFlow, Keras, OpenCV, scikit-learn, Hugging Face, Transformers, MLflow GenAI & LLMs LLM Evaluation, Prompt Engineering, RLHF, Model Alignment, HIL Systems, LangChain Multimodal EEG/BCI, MFCC, Video Transformers, Cross-modal Fusion, Attention Mechanisms Data & ETL Python, SQL, Pandas, NumPy, Selenium, BeautifulSoup, Neo4j, n8n Full Stack Node.js, TypeScript, React, PostgreSQL, Supabase, Tailwind CSS, Firebase, AWS Cognito MLOps AWS, Docker, Kubernetes, Git, Linux, REST APIs, CI/CD Security OAuth 2.0, SAML 2.0, SSO, Splunk, CrowdStrike, ServiceNow

Contact this candidate