Ansh Roshan
Open to Gen AI roles

Engineering AI products, end to end.

Agentic systems, RAG pipelines, and the products around them, shipped at TCS.

Currently building —

0+Projects built
0Shipped to production
0Toolchain domains
0AI specialisations
Ansh RoshanBengaluru

Ansh Roshan

AI Engineer @ TCS

Certified

Claude Architect · Pro

Certified

AWS · Solutions Architect

Google Cloud

Prompt Design · Vertex AI

agent-trace.log

$agent.ask("hybrid vs pure vector?")

→route bm25 + vector + graph

~recall 128 → rerank 24 → ctx 6

✓answer grounded · 4 citations

LangGraph · RAG · EvalsOpen source · MIT
01Stack

The stack I build with

Across Gen AI and the backend, frontend, and infrastructure that turn a model into a product.

LangChain
LangGraph
LlamaIndex
RAG pipelines
Agentic workflows
OpenAI
Anthropic Claude
Hugging Face
Vector databases
Python
TypeScript
Go
FastAPI
LangChain
LangGraph
LlamaIndex
RAG pipelines
Agentic workflows
OpenAI
Anthropic Claude
Hugging Face
Vector databases
Python
TypeScript
Go
FastAPI
Node.js
PostgreSQL
MongoDB
Redis
React
Next.js
Tailwind CSS
Framer Motion
Docker
Kubernetes
AWS
Terraform
Git
Node.js
PostgreSQL
MongoDB
Redis
React
Next.js
Tailwind CSS
Framer Motion
Docker
Kubernetes
AWS
Terraform
Git

Five domains, one stack

Core daily drivers

Gen AI & LLMs

09

Agentic systems, RAG pipelines, and LLM apps shipped to production.

LangChainLangGraphLlamaIndexRAG pipelinesAgentic workflowsOpenAIAnthropic ClaudeHugging FaceVector databases

Languages

03

The languages I reach for across AI, services, and tooling.

PythonTypeScriptGo

Backend & Data

05

APIs and data layers behind the models and the products.

FastAPINode.jsPostgreSQLMongoDBRedis

Frontend

04

Interfaces for AI products, built with motion and care.

ReactNext.jsTailwind CSSFramer Motion

DevOps & Cloud

05

Shipping and running it all, end to end.

DockerKubernetesAWSTerraformGit
02Selected work

Systems built to be judged in production.

Five systems that show how I think: retrieval, routing, evaluation, review tooling, and real-time collaboration.

All 17 projects
Loupe — AI PR Reviewer cover01 · Gen AI2026

Loupe — AI PR Reviewer

AI Developer Tooling

A self-built AI pull-request review agent in the spirit of CodeRabbit: a zero-dependency review engine that reads any diff, posts inline comments plus a summary, and runs as a GitHub Action or Cloudflare Worker with your own LLM key — no server required. Named for the jeweler's lens: close, careful inspection of every diff.

TypeScriptGitHub ActionsCloudflare Workers
Case study
OmniBench — LLM Benchmark cover02 · Gen AI2026

OmniBench — LLM Benchmark

LLM Evaluation

Point it at any model — API, aggregator, or local — and it runs a 5-pillar battery (capability, reliability, safety, agency, economics), recording every run, task, turn, span, and token. Compare models side-by-side in a full dashboard with CLI/TUI runners, three.js 3D data views, and portable run bundles that pool into community averages.

TypeScriptthree.jsCLI/TUI
Case study
RAGStack cover03 · Gen AI2026

RAGStack

RAG + Agents + Knowledge Systems

A hybrid agentic RAG system: it reads PDFs, Office docs, markdown, code, and web pages, builds four kinds of search indexes — BM25 vectorless, LanceDB vector, an LLM-extracted knowledge graph, and read-only text-to-SQL — and an agent decides which to chain for each question, always answering with citations. Local-first, MIT.

PythonBM25LanceDB
Case study
LLM Router cover04 · Gen AI2026

LLM Router

LLM Infrastructure + Optimization

A cache-aware LLM router with learned quality routing: one decision engine, two runtimes (pip + npm), zero dependencies, and an HTTP /decide sidecar that slots into LiteLLM and Bifrost to pick the best model per request on cost, latency, and quality.

PythonTypeScriptLiteLLM
Case study
Cortex cover05 · Full-stack2026

Cortex

Real-time Collaborative Editor

A private, authenticated, multi-file collaborative workspace that presents as an AI workspace on the surface — behind the login it is a real-time collaborative code editor built on Monaco and operational transforms, with a Rust + WASM backend in a single Docker image.

RustWASMMonaco
Case study
Next step

Let's build something that holds up in production.

Whether it's an agentic system that needs to stop hallucinating, a retrieval pipeline that needs to actually retrieve, or a product that needs shipping? Start with a message.

Start a conversationianshroshan@gmail.com
Replies within 24hIST · Bengaluru