chitraksh.
Resume

AI Engineer at Orion Innovation, Chennai, India

SingisettiChitraksh

I build production GenAI systems, and the evidence that they work.

  • RAG systems
  • agentic workflows
  • LLM evaluation
  • vector search
  • prompt engineering

I'm an AI Engineer at Orion Innovation in Chennai, working there since November 2024. I build retrieval-augmented generation pipelines, agentic workflows and LLM evaluation for enterprise clients, including a financial document intelligence system and an internal HR knowledge assistant. Outside work I build projects to test ideas end to end, such as a deep research agent, an LLM benchmarking framework and a hallucination detector.

Resume (PDF) Email me GitHub

↓ measurements, plates and a small search inside

Chapter I

Measurements

Numbers from my own work, taken from my resume.

  • 40%

    less time for analysts to find answers in financial documents

    Orion, client project
  • 35%

    fewer hallucinations after prompt and grounding work

    Orion, client project
  • 25%

    higher retrieval recall from the OpenSearch RAG pipeline

    Orion, client project
  • 10,000+

    documents ingested, chunked and indexed for Nexus

    Orion, Nexus
  • 40%

    faster query resolution with the HR assistant

    Orion, Nexus
  • 97%

    classification accuracy fusing RGB and hyperspectral features

    IEEE conference paper

Chapter II

Expeditions

Where I've worked, newest first.

  1. 01

    Nov 2024 to present

    Chennai, India

    Orion Innovation

    AI Engineer

    Building production GenAI systems for enterprise clients.

    GenAI financial document intelligence (client project)

    • Developed a GenAI document intelligence solution for financial document Q&A, cutting analyst document-search time by 40%.
    • Built end-to-end RAG pipelines with embeddings, semantic search and an OpenSearch vector database, improving retrieval recall by 25%.
    • Designed agentic workflows in LangGraph for multi-step orchestration, retrieval and tool integration, cutting manual follow-ups by 30%.
    • Applied prompt engineering, few-shot prompting and context grounding, improving accuracy by 20% and cutting hallucinations by 35%.
    • Built LLM evaluation workflows for relevance, recall, faithfulness and consistency, raising faithfulness by 15%.
    • Used AWS S3, EC2 and Lambda for scalable document processing and LLM inference, lowering latency by 25%.
    • RAG
    • LangGraph
    • OpenSearch
    • LLM evaluation
    • AWS

    Nexus, an enterprise GenAI knowledge assistant

    • Developed OI Nexus, an internal AI chatbot for employee questions on HR policies, hybrid work and leave, cutting query resolution time by 40% and repetitive HR tickets by 30%.
    • Built and optimized RAG pipelines with embeddings and vector search over enterprise document repositories, improving retrieval accuracy by 25% and reducing irrelevant results by 35%.
    • Engineered document ingestion, chunking and indexing workflows for large internal knowledge bases, processing 10,000+ documents while cutting indexing latency by 20%.
    • Refined response quality through prompt engineering, retrieval tuning and evaluation, raising answer accuracy by 15% and user satisfaction by 20%.
    • Partnered with HR and product stakeholders to define requirements and ship features, driving a 30% adoption increase in the first quarter.
    • RAG
    • Vector search
    • Prompt engineering
    • Chatbot
  2. 02

    Jul 2024 to Nov 2024

    Bangalore, India

    Liquidmind AI

    Software Engineer Intern

    Backend and document-automation work for business workflows.

    WhatsApp automation and OCR

    • Developed and deployed backend Flask APIs to integrate a WhatsApp bot for business automation workflows, reducing manual message-handling effort by 30%.
    • Built an OCR pipeline using Document AI to extract structured information from unstructured documents, improving extraction accuracy and cutting manual data-entry time by 40%.
    • Worked with senior engineers to test and refine API endpoints before release, improving reliability and reducing post-deployment bug reports.
    • Flask
    • REST APIs
    • Document AI
    • OCR
    • WhatsApp bot

Schooling

Education

B.Tech, Computer and Communication Engineering

Amrita Vishwa Vidyapeetham, Coimbatore, 2020 to 2024

Ribbons and badges

Certification

AWS Certified Cloud Practitioner

Amazon Web Services

Publication

Plant Disease Detection Using Feature Fusion

IEEE conference paper

Chapter III

Specimens

Projects I built to test ideas end to end. Open a plate for the details.

Plate 01

MicroManus

Investigator visibilis

Deep research AI agent

A full-stack deep research agent with a visible reasoning loop, so users see the model's multi-step research and tool use as it happens. Built without LangChain or LangGraph, and live.

read the plate
  • Added social login with OAuth and usage-based billing using Supabase and Stripe, for production-ready onboarding and metered monetization.
  • Designed a bring-your-own-key system that encrypts user-provided API keys with AES-256-GCM, giving self-managed access to multiple LLM providers.
  • Integrated SerpAPI for live web search grounding and deployed on Vercel as a public product.
  • Next.js
  • React
  • TypeScript
  • Node.js
  • Supabase
  • Stripe
  • SerpAPI
  • OpenAI API
  • Claude API
  • AES-256-GCM
  • Vercel

Live site

Plate 02

Plant Disease Detection Using Feature Fusion

Folium fusionis

Research, IEEE conference publication

A deep learning model that fuses spatial (RGB) and spectral (hyperspectral) features to classify plant diseases. It reached 97% accuracy, 5% better than single-modality baselines.

read the plate
  • Processed hyperspectral image data with band selection, normalization and feature extraction, and augmented 5,000+ images with OpenCV and NumPy, reducing overfitting by 15%.
  • Fused handcrafted HOG descriptors with ResNet50 deep features, improving robustness under varying lighting.
  • Co-authored the findings for an IEEE conference, contributing multi-modal fusion techniques to agricultural disease detection.
  • Python
  • TensorFlow
  • Keras
  • OpenCV
  • NumPy
  • Pandas
  • Scikit-learn

Plate 03

DualMind AI

Iudex duplex

LLM benchmarking framework

Compares Qwen2.5 and Claude Sonnet 4 on hallucination rate, bias resistance and content safety, using a 50-prompt test suite. FastAPI backend, GitHub Actions CI, deployed to Vercel, Render and Hugging Face Spaces.

  • Python
  • FastAPI
  • GitHub Actions
  • Vercel
  • Render
  • Hugging Face Spaces

Source

Plate 04

Grounding Auditor

Veritas atomica

Hallucination detection kit

Checks RAG output for hallucinations. It splits an answer into atomic claims, checks each claim against its sources and combines the results into a hallucination score. Built for the Lamatic AgentKit challenge with a Next.js interface.

  • Next.js
  • Lamatic AgentKit
  • GraphQL

Plate 05

NSE Stock Screener

Taurus ursusque

Real-time market analysis

A real-time screener for NSE-listed stocks, connected to Angel One's SmartAPI. It computes SMMA indicators, detects crossover signals, adds an ML scoring layer and shows everything in a Streamlit dashboard. Packaged as a Windows executable.

  • Python
  • Streamlit
  • SmartAPI
  • Machine learning

Still growing

Greenhouse

growing

Visa Advisor

Viator consultans

Tool-calling chatbot

Suggests countries and visa routes based on your interests, then walks through the application process end to end. Uses a Groq-hosted open LLM with SerpAPI tool calling for current web results.

  • Groq
  • SerpAPI
  • Tool calling
  • Vercel

Chapter IV

Try it

A small retrieval experiment: ask a question and it finds the matching passages on this page.

This is retrieval, not a chatbot. It ranks passages from this page with BM25 keyword matching, in your browser. No model is called and nothing is sent anywhere.

Chapter V

Toolkit

What I reach for.

Languages

  • Python
  • C

ML and deep learning

  • TensorFlow
  • Keras
  • PyTorch
  • Scikit-learn
  • Pandas
  • NumPy
  • Hugging Face Transformers

GenAI and LLMs

  • RAG
  • LangChain
  • LangGraph
  • LlamaIndex
  • OpenAI API
  • Anthropic Claude API
  • Prompt engineering
  • LLM evaluation (RAGAS)
  • Fine-tuning (LoRA, QLoRA)
  • vLLM
  • Ollama

Agentic AI

  • AI agents
  • MCP
  • Function and tool calling
  • Structured outputs
  • Model context engineering

Data and vector databases

  • FAISS
  • Chroma DB
  • Pinecone
  • OpenSearch
  • MySQL

Cloud and DevOps

  • AWS
  • Docker
  • Kubernetes
  • CI/CD
  • GitHub Actions
  • FastAPI
  • Git
  • Jupyter

Chapter VI

Send word

Open to conversations about AI engineering, contract work and side projects.

Resume (PDF) Email me GitHub