Ankit Kaushik — Full-Stack & Gen AI Engineer

Selected WorkFeatured ProjectsCase StudiesLive BuildsFull-Stack AppsShipped ProductsMy PortfolioExplore BelowSelected WorkFeatured ProjectsCase StudiesLive BuildsFull-Stack AppsShipped ProductsMy PortfolioExplore Below

Cost-Aware Model Routing Gateway for LLM Applications

LLMRouter

A routing gateway that scores query complexity with local embeddings and dispatches each request to the cheapest capable model in a tiered pool, escalating only ambiguous queries to an LLM-based judge so classification stays off the hot path. Hardened for production with per-request budget ceilings, automatic failover on provider timeouts and rate limits, and structured logging of every routing decision.

Python
LLM Routing
Embeddings
FAISS
Evaluation Harness
Cost Benchmarking
Structured Logging
Failover

View Code01/04

github.com/DeveloperAnkitKaushik/LLMRouter

Screenshot pending

Semantic Caching Layer for LLM Applications

SemCache

An embedding-based caching layer that matches semantically equivalent user queries — paraphrases, reworded prompts — to cached LLM responses, cutting redundant inference calls by ~55% on the eval workload. Similarity-threshold sweeps across ~1,200 query pairs pinned the operating point that keeps false-positive cache hits below 2%.

Python
FAISS
Ollama
Embeddings
Semantic Caching
Vector Search
Threshold Tuning

View Code02/04

github.com/DeveloperAnkitKaushik/SemCache

Screenshot pending

AI-Assisted Shopping Layer — Shipped at Wallmantra

Grace AI

An AI shopping assistant that lets customers find products in natural language and get personalised recommendations across the catalog. Shipped alongside LLM-powered support workflows that handle common queries automatically and keep responses consistent, cutting manual support load.

Next.js
Node.js
LLM Integration
RAG
Recommendations
Prompt Engineering
Caching

Live Demo03/04

wallmantra.com

Screenshot pending

Developer Portfolio Website

Portfolio

A production-grade personal portfolio built with Next.js, TypeScript, and Tailwind CSS, combining a monochrome design system with Lenis smooth scrolling and GSAP-driven motion — a scroll-pinned project showcase, a magnetic tech grid, and a themed circular wipe between light and dark.

Next.js
React
TypeScript
Tailwind CSS
GSAP
Lenis
React Bits

Live Demo04/04

kaushikankit.vercel.app

Screenshot pending

My Tech Stack

What I
Use

I utilize a comprehensive suite of modern technologies to build robust, scalable, and high-performance digital solutions.

My Journey

Work
Experience

A timeline of my professional experience shipping production LLM features — AI agents, retrieval pipelines, and workflow automation — alongside the platform work that keeps them fast and reliable.

May 2025 – Present
Software Engineer
Wallmantra • New Delhi
- Built and shipped "Grace AI", an AI-assisted shopping layer that lets customers find products in natural language and get personalised recommendations across the catalog.
- Added LLM-powered customer support workflows that handle common queries automatically and keep responses consistent, reducing manual support load.
- Improved storefront performance, cutting page load times by ~25% through caching and query optimisation.
- Set up a modular, reusable service architecture with onboarding docs that helped new developers get productive faster.
- Worked with product, design, and growth teams to ship AI features that made products easier to discover and improved checkout.
Mar 2025 – May 2025
Software Engineer Intern
Amantya Technologies • Onsite, Gurugram
- Designed multi-step agent workflows that automated client request handling and reduced manual work for the operations team by ~40%.
- Built a structured logging and monitoring setup with automatic file rotation that significantly cut log storage and made issues easier to trace.
- Standardised log formats and error-tracing across services, making debugging and deployments faster and more reliable.