Hi, I'm Rishabh
AI Full-Stack Engineer · Final-year B.Tech (IT) @ IIIT Allahabad. Applied AI systems engineering — from model design through agent orchestration to user-facing apps.
RK

About

Final-year B.Tech (IT) at IIIT Allahabad, working at the intersection of applied AI and shipped product. Most recently completed a Technology Internship with ICICI Bank's Monitoring & Observability team, contributing to a banking-grade AIOps platform with ML-based anomaly detection. Currently focused on AI/ML research at IIIT Allahabad — including a 5-tier edge-deployed CBC analysis pipeline on Jetson — alongside applied LLM systems being prepared for release.

Work Experience

Skills

Python
PyTorchPyTorch
TensorFlowTensorFlow
Hugging FaceHugging Face
LangChainLangChain
OpenAIOpenAI
AnthropicAnthropic
Google GeminiGemini
FastAPIFastAPI
Next.js
React
Typescript
Node.js
PostgreSQL
Docker
Amazon Web ServicesAWS
My Projects

Check out my latest work

I've worked on a variety of projects, from simple websites to complex web applications. Here are a few of my favorites.

financebench-rag-agent

Multi-agent LangGraph RAG system for financial document Q&A, reaching 47.3% correctness on the FinanceBench benchmark (150 questions, 32 companies) — within 8 points of the EMNLP 2025 SOTA — via selective agentic retrieval and a LoRA fine-tuned BGE reranker. RBAC enforced at the Qdrant payload-filter layer, HITL approval on high-stakes answers via LangGraph interrupt(), and a self-hosted LiteLLM + Langfuse v3 + Redis observability stack. Backend complete with 294 passing tests; Next.js 16 frontend in progress.

LangGraph
FastAPI
Qdrant
Claude Sonnet 4.6
PostgreSQL
Next.js (WIP)

edge-cbc-analyzer

5-tier Complete Blood Count analysis pipeline (segmentation → WBC differential → anemia screen → RBC morphology → 15-target CBC regression) on Jetson Orin Nano Super + TX2. YOLOv8n-seg matches published SOTA at mAP50 = 0.988 with 2.7–3.3× fewer parameters; end-to-end 171 ms / 5.85 FPS; 30-min stress run with no thermal throttling or memory leak.

PyTorch
TensorRT FP16
YOLOv8
EfficientNet-B0
Jetson
Gradio

zen-daily-wisdom

Personalized RAG service over 4,606 wisdom passages, with bandit-based mood personalization (regret 47.1 vs 124.8 baseline) and RAG recall@5 = 0.767. Multi-channel delivery via email/Telegram, deployed end-to-end on Vercel + Northflank.

Next.js
FastAPI
Supabase pgvector
Gemini
Vercel

banking-observability-platform

Open-source AIOps stack for 37+ banking microservices with ML-based DDoS detection (Isolation Forest, LSTM, One-Class SVM), a chaos-engineering injector, and a Mermaid-mapped architecture across 25+ containers.

Python
Docker
Kubernetes
TensorFlow
Prometheus
Grafana

skywings-airways

Full-stack airline ticket booking with real-time Google Flights data, OTP-secured mock payments, and PDF/QR e-tickets. Deployed on Vercel + Oracle Cloud with 55 passing unit and integration tests.

Spring Boot 3
React 18
PostgreSQL
Redis
Java 21

amazon-greencart-hackathon

Sustainable e-commerce platform with an XGBoost-based EarthScore (0–100), DBSCAN-driven group-buying clusters, and a multi-agent LangGraph shopping assistant. Open-sourced and reached Semi-Finals as Team Dragonz; 58 ⭐ on GitHub.

React
TypeScript
FastAPI
LangGraph
Redis
Gemini
Hackathons

Hackathons

I've participated in a few hackathons over the past year. Notable mentions below.

Amazon HackOn — Season 5

Online · India

Built GreenCart — a sustainable e-commerce platform with an XGBoost-based EarthScore, DBSCAN-driven group buying, and a multi-agent LangGraph shopping assistant. Open-sourced — 58⭐ on GitHub. Reached Semi-Finals as Team Dragonz.

Walmart Sparkathon — Converge

Online · India

Walmart Global Tech's open innovation challenge. Submitted GreenCart as a sustainability-focused commerce concept; recognised with a Certificate of Appreciation from Walmart Global Tech.

Contact

Get in Touch

Open to engineering opportunities. Reach out on LinkedIn or via email to discuss AI systems, agentic architectures, or applied ML.

GitHub
LinkedIn