AI / ML Engineer

Anurag
Pandey.

I build production-grade generative AI — multi-agent orchestration, retrieval-augmented generation, and computer vision systems that ship to real users.

View selected work Get in touch
Scroll
01

About

I'm an AI/ML engineer focused on taking generative AI from notebook to production — building systems that are accurate, safe, and genuinely useful.

My work spans multi-agent orchestration, retrieval-augmented generation, and computer vision, built with Python, PyTorch, LangGraph, FastAPI and AWS. I've architected a medical-imaging platform that classifies MRI and CT scans at 94% accuracy, scaled an enterprise RAG system to 500+ daily queries, and engineered a privacy-first natural-language copilot with self-repairing query loops and zero row-level data leakage.

Whether I'm fine-tuning domain-specific language models, designing secure query-validation frameworks, or shipping real-time analytics, I care about the full path from research to a reliable system people can trust.

94%
Diagnostic accuracy on MRI / CT imaging
500+
Daily queries served by enterprise RAG
95%
Intent-classification accuracy, agentic copilot
60%
Reduction in diagnostic latency
02

Experience

Where I've shipped.

  1. Apr 2026 — Present

    AI Engineer LLM Systems & Agentic AI

    Opsight AI Pvt Ltd

    • Architected a privacy-first AI copilot with LangGraph multi-agent orchestration across enterprise workflows — lifting intent-classification accuracy to 95% with zero row-level data leakage.
    • Engineered an offline time-series analytics engine (Pandas, SciPy, statsmodels) computing performance aggregates and flagging anomalies via linear regression and Median Absolute Deviation.
    • Designed a secure query-validation framework with regex allowlisting and schema grounding to eliminate Flux-injection vulnerabilities, verified by a full pytest integration suite.
  2. Nov 2025 — Jan 2026

    AI/ML Systems Engineering Intern

    Engagesphere Technology Pvt Ltd

    • Built a RAG banking explorer (LangChain, Pinecone) serving 500+ daily queries at 90% precision through context-chunking.
    • Deployed a fine-tuned BERT classification pipeline for anomaly detection, streamlining compliance monitoring across 3 departments.
    • Programmed a custom PII data-masking module using Named Entity Recognition, with interactive Plotly dashboards generating 40 audit-ready compliance reports.
  3. May 2024 — Jul 2024

    Healthcare AI Research Intern

    AIIMS, Raipur

    • Fine-tuned a domain-specific BERT variant on genomic literature, improving context-retrieval accuracy by 24% for clinical knowledge extraction.
    • Designed a LangChain agentic system with stateful prompts and sequential routing, enabling 50+ researchers to query complex genetic findings.
    • Built structured parsing with custom serialization schemas, delivering publication-ready summaries that cut clinical review time by 60%.
03

Selected work

Problem → approach → impact.

JeevVeda

AI-Powered Cancer Detection Platform Jun 2025 — Present

Problem

Cancer diagnosis is bottlenecked by slow, manual review of imaging and blood reports — delaying triage when speed matters most.

Approach

A ResNet50 CNN classifies MRI/CT scans, paired with an OCR blood-report analyzer on the Gemini API and a RAG layer for clinical context, served through a FastAPI + Next.js stack.

  • 94% accuracy (0.91 AUC) on MRI/CT classification
  • −40% biomarker review time via OCR analyzer
  • −60% diagnostic latency for faster patient triage
  • Python
  • ResNet50
  • TensorFlow
  • Gemini API
  • LangChain
  • RAG
  • FastAPI
  • Next.js
  • MongoDB
Live demo Source

MedBuddy

AI Medical Document Assistant Dec 2025 — Jan 2026

Problem

Clinical teams lose hours manually searching medical records and documents for a single answer.

Approach

A RAG chatbot over FAISS vector search ingests PDFs with recursive chunking and Hugging Face embeddings, returning grounded answers in seconds.

  • 88% answer relevance across 200 documents
  • <5s response time per query
  • −80% manual review effort for research workflows
  • Python
  • Streamlit
  • RAG
  • FAISS
  • LangChain
  • Hugging Face
  • Sentence-Transformers
Source

BankDataLens

AI Banking Data Explorer Nov 2025 — Dec 2025

Problem

Financial stakeholders can't interrogate transaction data without analysts and hand-written SQL.

Approach

A natural-language RAG explorer (LangChain + Pinecone) with an ETL pipeline, Isolation-Forest anomaly detection, and live Plotly dashboards.

  • 92% semantic precision on financial queries
  • 500 transactions scanned for anomalies
  • −75% analysis time via NL querying
  • Python
  • Streamlit
  • RAG
  • Pinecone
  • LangChain
  • Plotly
  • Pandas
  • Scikit-learn
Source
04

Skills

The toolkit, by domain.

Languages, Backend & UI
  • Python
  • SQL
  • FastAPI
  • React 19
  • Streamlit
  • Gradio
  • Pydantic v2
Generative AI & Agentic Systems
  • LangGraph
  • LangChain
  • CrewAI
  • Hugging Face
  • Transformers
  • Prompt Engineering
LLM Fine-Tuning & Optimization
  • PEFT
  • SFT
  • LoRA
  • QLoRA
  • Unsloth
  • Weights & Biases
Deep Learning & Data Science
  • PyTorch
  • TensorFlow
  • Pandas
  • NumPy
  • SciPy
  • statsmodels
  • Scikit-learn
Databases & Vector Stores
  • InfluxDB (Flux)
  • Pinecone
  • FAISS
  • ChromaDB
  • MongoDB
Cloud, MLOps & DevOps
  • AWS EC2
  • AWS S3
  • Docker Compose
  • GitHub Actions
  • Git
  • uv
  • pytest
ML & Math Foundations
  • Linear Algebra
  • Probability & Statistics
  • Calculus
  • Optimization
  • Hypothesis Testing
05

Education

2022 — 2026

B.Tech

National Institute of Technology (NIT), Raipur

CGPA 7.94/10
2021

CBSE Class XII

M.T.S Public School, Mirzapur

87.6%
2019

CBSE Class X

M.T.S Public School, Mirzapur

89.6%
06

Recognition

Achievements & certifications.

Achievements

  • 1st

    First Prize — Vigyaan Aavartan 2023

    Built an AI-powered robot with real-time object recognition; ranked first among 500+ participants.

  • 2nd

    1st Runner-Up — Technex, IIT (BHU)

    National robotics competition, outperforming 1000+ participants.

  • ₹50K

    Grant — DST-NIDHI TBI

    Secured funding for an AI-powered early cancer-detection startup in Central India.

Certifications

  • Generative AI with NLP, Agentic AI & Fine-Tuning Jan 2026
  • RAG Masters Dec 2025
  • Full-Stack Data Science May 2025
  • SQL: Beginner to Advanced for Data Professionals Nov 2024
07

Have a project or role
in mind? Let's talk.

I'm always glad to discuss generative-AI systems, RAG, agents and applied ML. The fastest way to reach me is email.

Email me