Abhinav
Nallamalli

Built a university-exclusive marketplace that replaces fragmented and untrusted resale channels with a verified, structured commerce system, enabling students to safely buy and sell within a closed university network. Designed end-to-end product flows across listing lifecycle, search and discovery, payments, fulfillment, and returns, while introducing trust layers such as user verification, transaction history, and a 24-hour return mechanism to improve reliability and user confidence. Engineered a scalable data model and pipelines to capture user behavior, transaction events, and item lifecycle data including reuse tracking, enabling insights into engagement, pricing trends, and sustainability metrics. Leveraged these data signals to optimize product discovery, reduce ghost listings, and improve transaction efficiency, while building a foundation for future ML-driven recommendations and pricing intelligence. Positioned the platform as a scalable, multi-university ecosystem with monetization through transaction fees, featured listings, and data-driven insights.

Built a HIPAA-compliant AI health assistant that helps users understand complex medical bills and insurance documents by allowing them to upload files and ask questions in natural language, solving the problem of confusion and lack of transparency in healthcare costs. Developed ML-driven pipelines for EOB parsing, CPT/ICD-10 mapping, and cost prediction, achieving around 70% accuracy in document understanding and 65% precision in cost estimation while reducing billing confusion by 35%. Architected a scalable cloud backend using AWS Lambda, S3, EC2, and Snowflake to integrate CMS APIs, EHR data, and insurance systems for real-time insights. Led end-to-end product development by defining user flows and building a conversational AI experience that translates complex healthcare data into clear, actionable decisions for users.

A context-aware AI system that interprets human emotion from spoken words, vocal tone, and facial micro-expressions — mimicking how humans perceive sentiment beyond text. Built a 3D CNN visual pipeline to capture temporal cues from video sequences, and engineered acoustic feature extraction via OpenSMILE (MFCCs, pitch, jitter, energy contours). A cross-modal attention framework lets text, audio, and visual streams refine each other's predictions, improving accuracy from 45% → 58% and F1 from 0.72 → 0.85. Achieved a 70% improvement over unimodal baselines and reduced inference latency from 8s → 5s per sample.

Built FactoryVoice, an AI-powered conversational interface that enables factory teams to access operational insights using natural language instead of relying on dashboards or SQL queries. The product was designed to solve delays in decision-making caused by fragmented data access, allowing users to quickly analyze demand, on-time delivery, and root causes of production issues. It processes large-scale manufacturing data and delivers real-time, actionable insights tailored to non-technical users. The system intelligently routes queries through semantic matching or dynamically generates SQL to retrieve relevant data, making complex analytics accessible and intuitive. This significantly reduced the time to retrieve insights from hours to seconds and improved operational efficiency across teams.

Built a Driver Risk Analytics System to proactively identify high-risk drivers using large-scale telematics data, enabling operations and compliance teams to move from reactive monitoring to data-driven intervention. Designed and implemented a distributed data pipeline using HDFS, Hive, and Impala to process geolocation, driver logs, and sensor data at scale, ensuring consistent and reliable analysis across drivers, vehicles, and regions. Developed a risk scoring framework by normalizing behavioral signals such as speeding frequency, mileage, and event severity, allowing teams to prioritize high-risk drivers and take targeted action. Generated insights across dimensions like city-level risk clusters and vehicle performance, helping inform routing strategies, safety interventions, and insurance decisions. Delivered interactive dashboards that translated complex data into actionable insights, improving visibility into fleet risk and supporting ongoing operational optimization.

Built a context-aware content discovery platform from scratch featuring SceneScout, a multimodal search system that allows users to upload clips, images, or describe scenes to instantly identify and discover relevant content using computer vision and NLP. Designed additional features like mood-based recommendations, time-aware suggestions, and personalized recap summaries to reduce decision fatigue and improve engagement through intent-driven discovery. Engineered a scalable ML pipeline combining embedding-based similarity search, user behavior modeling, and ranking systems, while developing a Swift-based front-end prototype to simulate real-time user interaction flows. Focused on bridging product intuition with full-stack development, translating user intent into intelligent recommendations through end-to-end system design.