Ankit Kumar alsoankit

⚡ ANKIT KUMAR

Software Developer · Full-Stack · AI Systems

⚡ Build → Break → Understand → Improve

I like taking an idea from a rough concept to something that actually runs.

That means working across the frontend, APIs, data layer, integrations, deployment, and the problems that show up once everything is connected.

I'm especially interested in AI systems where the engineering goes beyond an LLM call: retrieval, evidence, evaluation, and figuring out why a system fails.

🚀 Things I've Built

🧠 Anchor

Grounded RAG Engine

A retrieval and verification pipeline designed to keep generated answers tied to source evidence.

The interesting part

Semantic retrieval alone missed information hidden behind document cross-references. Anchor follows those references, gates weak evidence, and verifies claims in the final answer.

`70% → 100%`

Multi-hop retrieval after adding citation following.

Python FastAPI PostgreSQL pgvector LLMs

View Anchor →

The idea

Retrieval decides what the model is allowed to see.

Verification decides what it is allowed to say.

Retrieval

Semantic search
Citation following

Verification

Evidence gating
Claim checking

💸 Splitora

Group expense splitting

A full-stack expense platform with live dashboards, debt simplification, OCR scanning, and AI insights.

Next.js TypeScript AI

View Splitora →

What I worked on

Frontend
Product UI, live dashboard, animations and user flows.

Integration
Integrated the application's AI features and connected the different application components.

Deployment
Production deployment and the services behind the AI features.

📊 GitHub Activity

🧰 Toolkit

const stack = {
  // ⚡ Full-Stack
  fullStack: [
    "React",
    "Next.js",
    "TypeScript",
    "Node.js",
    "FastAPI",
    "PostgreSQL",
    "Docker"
  ],

  // 🧠 AI Systems
  aiSystems: [
    "RAG",
    "LLMs",
    "Embeddings",
    "pgvector",
    "Retrieval",
    "AI Evaluation"
  ]
};

🧠 How I Think About Building

IDEA  →  BUILD  →  SOMETHING BREAKS  →  FIGURE OUT WHY  →  MAKE IT BETTER

I prefer projects that leave me with a better mental model of how the system actually works.

The first version usually answers "can this work?"

The debugging usually answers the more interesting question: "why did it work that way?"

Navigation Menu