Rakesh Kumar — AI/ML Engineer

Who I Am

Closing the gap between research and production.

AI/ML Engineer pursuing an M.Tech in AI & ML at VIT Vellore, with a B.Tech in Computer Science from KIET. I specialize in building production-grade LLM systems — from agentic RAG pipelines to fine-tuned models deployed at scale — backed by hands-on experience in cloud infrastructure, distributed systems, and MLOps.

I care about closing the gap between research and production: writing systems that are accurate, reliable, and built to scale from day one. I'm also extending into enterprise AI domains — LangSmith, the OpenAI API, prompt evaluation, and observability.

Vellore, IN LLM Systems Focus Freelance / Full-time

8.38

GPA M.Tech AI/ML

555K+

URLs in Production

96%+

Model Accuracy

0.9932

Phishing AUC

Technical Stack

Tools of the trade, by domain.

Languages

Python C++ SQL

Developer Tools

PostgreSQL AWS MongoDB MySQL Kafka Redis Load Balancer Linux

ML / Deep Learning

NumPy Pandas Scikit-learn PyTorch TensorFlow CNNs RNNs / LSTMs / GRU Transformers Transfer Learning Fine-tuning GANs Object Detection (YOLO, Fast R-CNN)

NLP / GenAI

Generative AI NLP (NLG/NLU) LangChain LangGraph RAG FAISS Vector Search Embeddings Prompting

Large Language Models

LLM evaluation LLM finetunning LLM Gateway LLM Guardrails Prompt Caching vectorless Rag ↗ LLM Eval Notes

MLOps

Git / GitHub DVC MLflow GitHub Actions (CI/CD) Docker Kubernetes

System Design

Distributed Systems Load Balancing Caching CDN API Gateway Microservices Event-Driven Arch. Sharding WebSockets High Availability

CS Fundamentals

Data Structures & Algorithms DBMS Operating Systems OOP

Selected Works

Production systems, ML pipelines, and infrastructure I've designed and shipped.

Enterprise Multimodal RAG System

Multimodal RAG supporting text, PDFs, images, audio, video and source code using LangChain and LangGraph — Hybrid Retrieval (Dense + BM25), Cross-Encoder reranking, multimodal embeddings , and hierarchical memory via LangGraph SQLite Checkpointer + MongoDB. Evaluated with Ragas.

LangGraph Hybrid Retrieval ImageBind AWS ECS

Distributed Cloud Deployment Platform

V1: scalable Git-to-deployment platform with isolated Docker builds on AWS ECS and S3 storage, using a custom streaming reverse proxy to serve deployments without memory overhead. V2: added PostgreSQL/Prisma for deployment data, migrated logging from Redis to a Kafka + ClickHouse pipeline for persistent, real-time build log ingestion.

AWS ECS Fargate Kafka ClickHouse Next.js

Hybrid Phishing URL Detection

Stacked an SGD Classifier, DeepTCN + Attention (AUC: 0.9932), and Isolation Forest via a Random Forest meta-learner — 96%+ accuracy/recall on 555K+ URLs. Built an 80,008-dim sparse feature pipeline (TF-IDF char n-grams + lexical features) with CSR compression, cutting memory from 32GB to 200MB. Real-time concept drift monitoring (Page-Hinkley, ADWIN, DDM) with active learning for human-in-the-loop retraining.

PyTorch DeepTCN Isolation Forest Concept Drift

LLM Guardrails: Multi-Layer AI Safety Pipeline

Async, cost-tiered guardrail pipeline (validation → PII → ML models → LLM) inspecting prompts pre-inference with DeBERTa-v3 for prompt injection detection, Detoxify for multi-label toxicity classification, and Presidio + spaCy for PII detection, with a standardized result schema and short-circuit logic. Asyncio-threaded transformer inference under concurrent load; extensible for post-generation checks and a planned LLM-as-judge hallucination layer.

FastAPI DeBERTa-v3 Detoxify AsyncIO

Experience & Education

Where I've worked, learned, and given back.

Experience Log

Jul 2025 — Present

M.Tech, AI & ML

Vellore Institute of Technology (VIT), Vellore

GPA 8.38 / 10 — specializing in advanced machine learning, deep learning, and AI applications.

Jul 2023 — Oct 2023

Full Stack Web Developer Intern

Johnnette AI

Built full-stack apps with React.js, Node.js & Express.js. Implemented RESTful APIs with JWT auth and role-based access control. Created reusable React component libraries improving UI consistency across product modules, and developed secure session management systems.

Nov 2020 — Jul 2024

B.Tech, CS & IT

KIET Group of Institutions, Ghaziabad

CGPA 7.3 / 10 — foundation in computer science fundamentals, data structures, algorithms, and software engineering.

Ongoing

DSA Coordinator & Uddesya Club Member

Leadership & Community

Designed and facilitated group DSA workshops for junior students, improving their problem-solving approach and CP ratings. Coordinated community outreach for 50–200 beneficiaries across food distribution, blood donation, and educational programs.