---
title: "Ayush Kumar - Graduate Engineering Trainee in India | Tal"
description: "Meet Ayush Kumar, GET @ Ola | AIR-2, Engineers Ring of Honour | Codeforces Max Rating : 1727 | IIT Guwahati CSE ’26 based in India. AI stack includes claude code, claude, and codex. Explore public work links including github, codeforces, and redluigi1.github.io."
image: "https://tal.club/meet/ayush-kumar-67/share-image"
---

# Ayush Kumar on Tal

## Headline

GET @ Ola \| AIR\-2, Engineers Ring of Honour \| Codeforces Max Rating : 1727 \| IIT Guwahati CSE ’26

## About

I specialize in multi\-agent orchestration, agentic RAG pipelines, and applied machine learning systems\. I also enjoy algorithmic problem solving and am exceptionally good at leveraging LLMs to automate and accelerate workflows\.

## Location

India

## AI stack

claude code

claude

codex

hermes agent

langchain

langgraph

## Work experience

Graduate Engineering Trainee at Ola Electric, Jun 2026 \- Present · 5 mo, Bengaluru, Karnataka, India, I work on applied AI at Ola Electric, building LLM systems that run in production\. AI warranty adjudication pipeline I built a system that approves or rejects warranty claims against part\-specific SOPs and gets about 90% accuracy compared to human reviewers\. It closed a fraud loop where technicians were claiming replacements for healthy parts and reselling the originals, which cut daily claims from 1,500 to 500 and saves roughly $1M a month\. I also built a self\-improving loop\. Whenever a human auditor corrects the model, the pipeline generates rule and prompt patches on its own, and they get validated against a golden set and shadow traffic before going live\. That took accuracy from 86% to 90%\. SAFAI, our in\-house CRM Co\-built a Salesforce replacement in 10 days, keeping access to 5M\+ historical job cards\. I was the primary engineer for the warranty and insurance integration with Reliance, Acko and Ola\. It rolled out pan\-India to 4,000\+ technicians and now handles about 5,000 job cards and 25K parts a day\. MCP war\-room agent Built an agent with access to our repo, DB and issue sheet\. It buckets tickets by root cause, finds the bug and raises the fix PRs, and it fixed around 3,400 job cards\. The MCP has hardcoded row limits, timeouts and an audit trail, because raw DB access for an LLM is a bad idea\.

Undergraduate Student Researcher \| BTP \| Under Prof\. Ashish Anand at CSE Department, IITG, Aug 2025 \- May 2026 · 10 mos, • LLM\-Amplified Data Engineering: Overcame critical data scarcity in complex public policy documents by engineering an automated dataset extension pipeline\. Utilized Gemini 2\.5 Pro with optimized 10\-shot Chain\-of\-Thought prompting and explicit negative constraints to scale the training corpus by 450%\. This structural augmentation drove a 425% improvement in relation extraction accuracy for downstream neural parsers\. Co\-authored a resource paper submitted to LREC 2026\. • Multi\-Agent RAG Architecture: Engineered a novel, end\-to\-end agentic Retrieval\-Augmented Generation \(RAG\) framework for argument mining\. Orchestrated specialized AI agents \(Component, Relation, and Critic\) via a central state manager to iteratively extract and map logical structures\. • Hybrid Semantic\-Structural Knowledge Graph: Built a dual\-track retrieval system integrating a vector database \(ChromaDB\) for semantic search with a Deep Multilingual Rhetorical Structure Theory \(DMRST\) parser\. Modeled the logical topology of texts into a NetworkX graph using spectral similarity, enabling the system to retrieve examples for in\-context learning which are not only semantically sound, but rhetorically as well\.

Intern \| Ola Maps at Krutrim, May 2025 \- Jul 2025 · 3 mos, On\-site, • Applied NLP & Search Optimization: Productionized a spaCy\-based Named Entity Recognition \(NER\) address parser for autocomplete and forward geocoding use cases\. This live deployment drove a 2% improvement in top\-3 search accuracy\. Conducted comparative evaluations with alternative transformer models, including BERT\. • Machine Learning & ETA Architecture: Upgraded live historical ETA lookup systems by applying Bayesian Optimization via Optuna to aggressively tune hyperparameters\. Engineered cyclic temporal and geo\-spatial features, achieving 60% of rides within 10% ETA deviation\. • Geospatial Clustering & Low\-Data Regions: Executed targeted ROI experiments to enhance ETA predictions in tier\-3 cities\. Utilized DBSCAN to cluster location\-based services \(LBS\) and routing segments \(wayIds\) into dynamic sub\-city zones\. • Algorithmic Debugging & Edge\-Case Resolution: Corrected underlying dataset logic to resolve systematic ETA overestimations on complex, long\-segment highway networks\. Discovered and patched critical routing anomalies affecting newly constructed, large\-scale transit links\. • Data Exploration & Feature Engineering: Analyzed massive Electric Vehicle \(EV\) ping datasets, successfully validating their informational density and viability for foundational routing models\. Experimented with advanced bucketization techniques to improve the predictability of route segment popularity\.

## Education

Bachelor of Technology \- BTech, Computer Science, Indian Institute of Technology, Guwahati

## Public links

- github: https://github\.com/Redluigi1
- codeforces: https://codeforces\.com/profile/mahnameisayush
- redluigi1\.github\.io: https://redluigi1\.github\.io/

[View the original profile](https://tal.club/meet/ayush-kumar-67)

```json
{
  "@context": "https://schema.org",
  "@graph": [
    {
      "@type": "Person",
      "@id": "https://tal.club/meet/ayush-kumar-67#person",
      "name": "Ayush Kumar",
      "url": "https://tal.club/meet/ayush-kumar-67",
      "image": "https://tal.club/meet/ayush-kumar-67/profile-photo.jpg",
      "description": "Meet Ayush Kumar, GET @ Ola | AIR-2, Engineers Ring of Honour | Codeforces Max Rating : 1727 | IIT Guwahati CSE ’26 based in India. AI stack includes claude code, claude, and codex. Explore public work links including github, codeforces, and redluigi1.github.io.",
      "jobTitle": "Graduate Engineering Trainee",
      "worksFor": {
        "@type": "Organization",
        "name": "Ola Electric"
      },
      "hasOccupation": {
        "@type": "Occupation",
        "name": "Graduate Engineering Trainee"
      },
      "homeLocation": {
        "@type": "Place",
        "name": "India"
      },
      "alumniOf": [
        {
          "@type": "EducationalOrganization",
          "name": "Indian Institute of Technology, Guwahati"
        }
      ],
      "knowsAbout": [
        "claude code",
        "claude",
        "codex",
        "hermes agent",
        "langchain",
        "langgraph"
      ],
      "sameAs": [
        "https://github.com/Redluigi1",
        "https://codeforces.com/profile/mahnameisayush",
        "https://redluigi1.github.io/"
      ],
      "mainEntityOfPage": {
        "@id": "https://tal.club/meet/ayush-kumar-67#webpage"
      }
    },
    {
      "@type": "ProfilePage",
      "@id": "https://tal.club/meet/ayush-kumar-67#webpage",
      "url": "https://tal.club/meet/ayush-kumar-67",
      "name": "Ayush Kumar on Tal",
      "description": "Meet Ayush Kumar, GET @ Ola | AIR-2, Engineers Ring of Honour | Codeforces Max Rating : 1727 | IIT Guwahati CSE ’26 based in India. AI stack includes claude code, claude, and codex. Explore public work links including github, codeforces, and redluigi1.github.io.",
      "dateModified": "2026-11-04T12:21:05.507Z",
      "isPartOf": {
        "@id": "https://tal.club/#website"
      },
      "mainEntity": {
        "@id": "https://tal.club/meet/ayush-kumar-67#person"
      },
      "primaryImageOfPage": {
        "@type": "ImageObject",
        "url": "https://tal.club/meet/ayush-kumar-67/profile-photo.jpg",
        "contentUrl": "https://tal.club/meet/ayush-kumar-67/profile-photo.jpg",
        "caption": "Ayush Kumar profile photo on Tal"
      },
      "inLanguage": "en-IN"
    }
  ]
}
```