What does Ashutosh Kumar build as a Generative AI Engineer?

He builds RAG systems that ground model answers in real documents, multi-agent workflows orchestrated with LangGraph, multimodal applications on Gemini Vision, and the full-stack Next.js and Python products that wrap them — including PostGen, AI Book Explorer and GenAI Apps Generator.

What is Ashutosh Kumar's GenAI tech stack?

Python and TypeScript as core languages; LangGraph, LangChain and CopilotKit for agent orchestration; Google Gemini and OpenAI APIs for generation; embeddings and vector search for retrieval; and Next.js, React, Node.js, PostgreSQL and Streamlit for the surrounding product.

Is Ashutosh Kumar available for hire?

Yes. He is open to Generative AI engineering roles and selected freelance work. The fastest way to reach him is by email at ashu.kumarexam@gmail.com or through LinkedIn at linkedin.com/in/ashusnapx.

What is Ashutosh Kumar's engineering background?

He has solved over 1,000 data structures and algorithms problems, mentored more than 600 students as a Teaching Assistant at Coding Ninjas with a 4.8/5 rating, shipped frontend work at AI Caller.io, and now develops enterprise SAP applications at Tata Consultancy Services for a European client.

What certifications does Ashutosh Kumar hold?

He is a Claude Certified Developer — Foundations (Anthropic, 2026), covering production applications and agents built on the Claude API, Claude Code, custom tools and MCP servers. He has also completed Anthropic's Claude Code in Action and Introduction to Agent Skills courses, Kestra Fundamentals, and is an SAP Certified Associate — Security Administrator. Every credential links to its issuer for verification.

Has Ashutosh Kumar won any AI hackathons?

Yes. His team placed Rank 1 in a TCS AI Friday Season 2 weekly hackathon, qualified for the semi-finals and went on to the regionals, with Nexus Cold Chain AI — an agentic system for vaccine cold-chain safety built on three LangChain agents, Gmail-MCP tool calling over OAuth 2.0, Mem0 contextual memory, PageIndex retrieval, Braintrust LLM evaluations and Pydantic AI schema validation.

Portfolio / 2026 Edition

Open to Generative AI engineering roles

Ashutosh Kumar — Bengaluru, IN

Generative AI Engineer

Role: Generative AI Engineer
Based: Bengaluru, India · remote-friendly
Focus: RAG · agents · AI product
Now: Software Engineer, TCS
Recent: Rank 1 — TCS AI Friday
Status: Open to GenAI roles

I build LLM systemsthat survive contactwith real users.

Retrieval pipelines grounded in real documents, multi-agent workflows that take more than one step, and the full-stack products they ship inside. Currently building enterprise software at TCS and generative AI everywhere else.

See the work ↓Start a conversation ↗

Fig. 01 — Ashutosh Kumar

15+
Products shipped: 1,000+
DSA problems solved: 600+
Engineers mentored: 3
Years in production

RAGLangGraphLangChainGeminiOpenAIMCPVector SearchEmbeddingsAgentsVoice AILLM EvalsPrompt EngineeringNext.jsTypeScriptPythonReactNode.jsPostgreSQLTailwindStreamlit

01Approach

Anyone can call an API.
Shipping it is the job.

A demo that works on the happy path and a system people depend on are different pieces of engineering. Here is where the difference goes.

01
Retrieval & RAG
Chunking strategies, embeddings and hybrid search wired into a retrieval layer that grounds every answer in real source documents instead of guessing.
- Embeddings
- Hybrid search
- Re-ranking
- Chunking
02
Agents & Orchestration
Multi-step workflows built as explicit state machines with LangGraph — tool calling, structured outputs and human-in-the-loop checkpoints instead of one giant prompt.
- LangGraph
- Tool calling
- Structured output
- MCP
03
AI-Native Product
The model is a third of the work. I ship the streaming UI, the auth, the rate limits and the database that turn a demo notebook into something people can actually use.
- Next.js
- Streaming UI
- TypeScript
- Postgres
04
Evaluation & Reliability
Prompts get versioned, outputs get schemas, and behaviour gets checked against fixed cases — because 'it worked when I tried it' is not a release criterion.
- Prompt versioning
- Schema validation
- Regression cases

02Selected work

Things I built,
not things I read about.

Every project below is live or open source. Start with the generative AI set — that is the work I want to be judged on.

Full source on github.com/ashusnapx

03Recognition

Judged against
everyone else.

Internal work is easy to describe and hard to verify. These are the results someone else scored.

Rank 01

TCS AI Friday/Season 2/2026

Nexus Cold Chain AI

An agentic system for vaccine cold-chain safety: it predicts when a shipment is about to breach temperature, explains why, and drives the incident response — rather than just raising an alarm and leaving a human to work it out.

Stage 01
Rank 1
Weekly hackathon
Stage 02
Qualified
Semi-finals
Stage 03
Reached
Regionals

What we built

Multi-agent orchestration: Three specialised LangChain agents — IoT sensor analyser, root-cause analyser and risk-prediction analyser — each owning one job instead of one prompt owning all three.
Unified data schema: A single schema joining IoT sensor telemetry, vehicles, warehouses, weather and mapping data, so the agents reason over one consistent view of the shipment.
Risk prediction: An ML model classifying every shipment as Normal, Partial or Critical, which is what turns raw telemetry into a decision the response system can act on.
Incident response: On a breach the system returns remaining repair time, the nearest viable warehouse alternatives and a packaging remedy — a recommendation, not just a red flag.
Tool use over MCP: Gmail-MCP behind OAuth 2.0 so the agent dispatches incident notifications itself, using the Model Context Protocol rather than a bespoke integration.
Memory & retrieval: A conversational layer built on PageIndex for document retrieval and Mem0 for contextual memory across turns.
Evaluation & validation: Braintrust for LLM evaluations and Pydantic AI for schema-validated outputs — the difference between a demo and something you would run unattended.

Stack

LangChain
Multi-agent
MCP
Mem0
PageIndex
Braintrust
Pydantic AI
OAuth 2.0
IoT telemetry

Built with

Vishnu Shanker Sharma · Shashank N U · Namratha S · Sachi Motghare

Read the write-up ↗

04Certifications

Credentials, and
the link to check them.

5 licences and certifications from Anthropic, Kestra, SAP. Every one links to the issuer so you can verify it rather than take my word for it.

AnthropicAug 2026 — Aug 2027
Claude Certified Developer — Foundations
Designed for developers who can build, integrate and ship production applications and agents on Claude using the Claude API, Claude Code, custom tools and MCP servers.
- Claude API
- Claude Code
- Custom tools
- MCP servers
Verify ↗
AnthropicMar 2026
Claude Code in Action
Agentic development with Claude Code — driving multi-step engineering work through tool use and the Model Context Protocol.
- Claude Skills
- Model Context Protocol (MCP)
Verify ↗ID uc66j6vuqecs
AnthropicMar 2026
Introduction to Agent Skills
Building and packaging Agent Skills — the reusable capability units that extend what an agent can do.
- Agentic AI Development
- Claude Skills
Verify ↗ID pk474prkjtf8
Kestra
Kestra
Kestra Fundamentals
Declarative orchestration of data and workflow pipelines with Kestra.
- Kestra
- Workflow orchestration
Issuer-verifiedID YKXTMLod
SAPAug 2024
SAP Certified Associate — Security Administrator
Core SAP authorization and security concepts across S/4HANA Public and Private Edition, applied as a security administrator.
- SAP S/4HANA
- Authorization
- System security
Verify ↗

05Stack

Tools I reach for
without looking them up.

Model layer first, because that is where the interesting problems are. Everything under it exists to make the model layer usable.

Generative AI

The model layer

LLM prompt orchestration
Retrieval-Augmented Generation (RAG)
LangGraph
LangChain
CopilotKit
Multi-agent workflows
Function / tool calling
Structured outputs
Google Gemini API
Gemini Vision (multimodal)
Model Context Protocol (MCP)
Embeddings & vector search
Voice AI / speech agents
Mem0 (agent memory)
Braintrust (LLM evals)
Pydantic AI
PageIndex retrieval

Languages

Day-to-day

Python
TypeScript
JavaScript
C++
C
SQL
ABAP

AI Product & Frontend

Where the model meets a user

Next.js 16 (App Router)
React 19
Server Components
Streaming responses
Tailwind CSS
shadcn/ui
Framer Motion
Redux Toolkit
React Query
React Native

Backend & Data

What holds it up

Node.js
Express.js
Streamlit
PostgreSQL
Prisma
MongoDB
Supabase
Appwrite
Redis
REST API design

Foundations

The part that does not go stale

Data Structures & Algorithms
System design
NumPy
Pandas
Plotly
Git & GitHub
Vercel
SEO & Core Web Vitals

06Journey

Teaching algorithms,
then shipping AI.

Enterprise engineering by day, generative AI by conviction. Both halves show up in how I build.

JUN 2025
↓ PRESENT
Software Engineer (Core ABAP)
Tata Consultancy Services (TCS)
Full-time, Onsite
- Developing and maintaining enterprise-grade business applications for a European client (Bayer) using Core ABAP.
- Performed system analysis, debugging, testing, and defect resolution within SAP-based production systems.
- Translated business requirements into technical solutions aligned with enterprise standards and compliance practices.
- Worked in controlled production environments following structured change management and release processes.
- ABAP
- SAP
- Enterprise Applications
- System Analysis
- Debugging & Testing
DEC 2023
↓ JAN 2024
Frontend Developer
AI Caller.io
Internship, Remote
- Implemented CSV upload functionality to trigger bulk calling workflows.
- Migrated application from client-side rendering to server-side rendering for improved performance and SEO.
- Integrated Hashnode’s headless CMS for blog management and better search engine visibility.
- Improved call template forms with tooltips to enhance user experience and usability.
- Integrated Tolt for affiliate marketing and referral tracking.
- NextJS
- ReactJS
- Redux Toolkit
- React Query
- TypeScript
- Shadcn
- Tailwind CSS
FEB 2022
↓ OCT 2022
Teaching Assistant
Coding Ninjas
Internship, Remote
- Solved more than 1000 questions related to C++, Data Structures, and Algorithms.
- Mentored and assisted more than 600 students in resolving coding and logical queries.
- Achieved a student rating of 4.8/5, reflecting strong teaching and problem-solving skills.
- C++
- Data Structures
- Algorithms

07GitHub

Consistency,
not a highlight reel.

Everything here is read live from the GitHub API — repositories, language mix and a year of commits. Boring on any given day, which is exactly why it is worth showing.

—
Public repos: —
Contributions / yr: —
Years on GitHub: —
Followers

Commit activity — last 12 months

github.com/ashusnapx ↗

Loading contributions_

Language mix — by repository

Recent repositories

08Writing

Explaining it is
how I know I know it.

Technical writing on Hashnode and Medium, pulled live from both feeds. Mostly fundamentals — the things that stay true after the framework of the month is gone.

Follow alongHashnode ↗Medium ↗LinkedIn ↗

09Featured

Hackathons, meetups
and the rooms in between.

Selected posts from LinkedIn — what I built, who I built it with, and what I took away.

10Questions

The things people ask first.

Ashutosh Kumar (ashusnapx) is a Generative AI Engineer and full-stack developer based in India. He builds LLM-powered applications — retrieval-augmented generation pipelines, multi-agent workflows and the production web products they ship inside — and currently works as a Software Engineer at Tata Consultancy Services.