Practical models,
provable utility.
NovaX Research develops applied machine learning models, edge-computing architectures, and pedagogical dialogue systems that power our commercial ventures: ShelfOne Retail OS, AlphaGrad EdTech, and PulseFit.
Stress-Test Edge Model Latency
Core Research Tracks
Active engineering streams.
We prioritize applied machine intelligence that directly solves latency, hallucination, and connectivity challenges in emerging markets.
Low-Resource Edge Inference
Compressing catalog search, fuzzy item matching, and demand prediction models into sub-20MB quantized weights capable of executing on low-cost quad-core cashier laptops with 0% cloud dependence. Directly powers ShelfOne's 3-second offline checkout.
Socratic Dialogic Tutoring
Developing specialized cognitive prompting pipelines that enforce didactic restraint. Rather than dumping answers, the model formulates counter-questions aligned with NCERT pedagogy, guiding students toward self-derived mastery on AlphaGrad.
Deterministic Tax Verification
Building hybrid neuro-symbolic verification layers that validate numerical computations against Indian GST statutory rules. Guarantees 0% mathematical hallucination across complex interstate IGST and cess calculations.
Offline-First Peer State Synchronization
Researching Conflict-Free Replicated Data Types (CRDTs) tailored for multi-counter retail stores where local counter machines must reconcile inventory levels during intermittent local network partitions without database deadlock.
Vernacular OCR & Handwritten Bill Parsing
Lightweight computer vision models optimized for recognizing handwritten Hindi and English wholesale parchas (vendor invoices), converting messy invoices into structured digital purchase orders in under 5 seconds.
Longitudinal Student Concept Heatmaps
Bayesian knowledge tracing models that map a student's conceptual prerequisite tree, distinguishing between arithmetic slip-ups and fundamental conceptual gaps across CBSE Class 9 to 12 STEM curricula.
Collaborate with NovaX Labs.
We partner with academic researchers, university engineering departments, and applied builders passionate about real-world AI systems.