Arjun Yadav
🚀 Founder @Spoon · Building @dataknot · Undergraduate Researcher @TCET
Full-Stack · Applied AI · ML Systems
Welcome
I build production-grade AI and backend systems — designed around real constraints like concurrency, unpredictable input, and failure modes. I take on ambiguous problems and turn them into systems that are measurable, debuggable, and deployable.
Hire me
I'm currently available for hire.
Download my resume [pdf]
I can build production-grade AI/ML systems, real-time backends, and high-concurrency applications that scale under real load. I specialize in taking ambiguous technical problems and turning them into reliable, measurable systems.
Verified Impact — Spoon
- 40 min → 5 min — Average customer wait time
- 525 → 887 / hr — Order-handling throughput
- 6+ colleges — Live deployment footprint
- Patent + Paper — Co-authored with R&D Dean
Projects
- Spoon (Patent-backed) — A high-concurrency virtual queue system built for canteens and similar service environments. Handles ML-based wait-time prediction, adaptive polling, atomic slot allocation, and idempotent Razorpay transactions with refund-safety. The system was designed to stay stable under peak load without ambiguity in order handoff — deployed across 6+ colleges with documented throughput improvement. Published research paper and patent co-authored with Dr. Vinitkumar Dongre (R&D Dean).
- NeuroGuard — Real-time EEG anomaly detection system for neurological signal analysis. Uses a custom 1D-CNN pipeline with Grad-CAM and SHAP-style explainability to support clinical-style decisions. Covers seizures, Alzheimer's, Parkinson's, ADHD, stroke, and tumor-related patterns. MVP built at MumbaiHacks '25 (5,000+ participants) — selected as a finalist among top teams.
- DeepGuard — In-browser deepfake detection that runs at the point of media consumption. Uses a multi-model agentic architecture with consensus-based scoring, explainable verdicts, and graceful fallback when a detector is unavailable — the goal is a transparent media-trust layer, not a black-box label. Built at eRaksha Hackathon (IIT Delhi) — selected among top teams at this national-level competition.
- RescuNav-AI — Multi-agent rescue simulation over a graph-modeled 3D environment. Implements A* pathfinding, dynamic danger modeling, and iterative mission learning — agents improve strategy measurably across runs. Currently being validated for real-environment deployment within a college setting.
- Dataknot (In progress) — An India-focused semantic data quality control plane. Targets drift detection, root-cause tracing, semantic contract enforcement, and automated correction workflows — built to make data failures easier to find, explain, and prevent at the source rather than patch downstream.
Recognition
- Avishkar 2025 — Maharashtra inter-university research convention — represented TCET
- eRaksha Hackathon (IIT Delhi) — DeepGuard — selected among top teams at this national-level competition
- MumbaiHacks '25 — NeuroGuard finalist — 5,000+ participant event
- SIH '25 — College-level winner — Smart India Hackathon
- 2× Hackathon winner — Across competitive open events
- MHT-CET 2024 — 94th percentile · AIR 19,595 of 7.5 lakh candidates