Shailesh Upadhyay

I don't just prototype models; I ship them into production. AI Backend Engineer with hands-on experience building conversational AI pipelines, high-concurrency backend services, and scalable web applications that don't break at 3 AM. From asynchronous FastAPI streaming and database query optimization to deploying sub-15ms inference engines, I build resilient systems that resolve core bottlenecks.

Experienced in shipping production Next.js web apps and client dashboards, and collaborating directly with US engineering teams to maintain 99.9% service reliability.

AI Backend Engineer at WebVory (2025-06 – Present) Core AI Infrastructure and Backend Systems • Collaborated with a US-based engineering team to build asynchronous FastAPI services handling concurrent LLM streaming calls with token rate-limiting and connection pooling, reducing p95 latency by 45% (1.2s → 660ms).

• Engineered the internal company-standard telemetry dashboard using React, TypeScript, and FastAPI, backed by PostgreSQL and Redis caching; cut median data query latency by 60%.
• Architected the complete GitHub Actions CI/CD pipeline (deploy.yml) with automated testing, linting, and branch validation gates, eliminating merge bottlenecks and maintaining 99.9% service uptime.

Freelance

Full-Stack and AI Developer at Freelance (2025-01 – Present) Client Engagements — Web Applications and Custom Dashboards • Architected the production web platform for Devine Child Development Centre and student dashboard app using Next.js, React, and TypeScript; deployed responsive UI workflows and optimized Core Web Vitals.

• Built an emotion-aware interactive system for an international client (LÜQ Prototype), implementing a 7-emotion sentiment classification engine with dynamic visual mood reactions. ML Engineer (Contract) at FAR Labs / Dizzaract (2025-12 – 2026-02) Player Behavior Cloning for Game AI • Trained 4 neural models (Aim LSTM, Movement Transformer, Tactical MLP) on 30Hz player telemetry (float[256] → float[32]) to clone movement and combat decisions.
• Implemented an Inverse Dynamics model to pseudo-label 500+ hours of gameplay; exported to ONNX and TensorRT (INT8/FP16) for sub-15ms production inference. NLP Engineer (Freelance / Client Project) at IOTA (2024-02 – 2024-11) Address Extraction NER Pipeline and Model Delivery • Engineered an end-to-end SpaCy NER pipeline with clean modular architecture (generation → training → evaluation → inference) to extract 4 core entity types (STREET, CITY, STATE, ZIP) from unstructured address strings.
• Developed a programmatic weak-supervision generator expanding 100+ raw inputs into 2,500 labeled training samples; trained model with SGD (50 epochs, 25% dropout), achieving P: 89.2% / R: 85.1% / F1: 87.1% on held-out test data for client handoff.

B.E. in Computer Science Engineering (AI/ML Specialization) – Chandigarh University (2022 – 2026)