Dubai, UAE · Immediately Available · Open to Relocate within UAE

Santosh Muruggu

AI & GenAI Engineer

I build production-grade LLM systems — RAG, multi-agent pipelines, and computer vision — with the evaluation rigor to prove they work.

View Projects Download Résumé Contact →

About

Engineer, not just a model-caller

AI/ML Engineer specializing in Generative AI and LLM applications, with hands-on experience designing and deploying retrieval-augmented generation (RAG) systems, autonomous multi-agent pipelines (LangGraph), and computer vision models. Shipped two production GenAI systems with rigorous evaluation harnesses (deepeval, LLM-as-judge) and full observability via Langfuse tracing. Strong foundation in Python, deep learning, and full-stack deployment (FastAPI, Streamlit, Docker). Co-authored ML research on constraint-driven PCB autorouting. Seeking an AI/ML Engineer or GenAI Engineer role in the UAE.

The portrait alongside is a code-generated point-cloud abstraction — sampled from a source photo at build time, posterized, and sparsified so it reads as data rather than a photograph.

Capabilities

Skills

Grouped by where they show up in production, not in a word cloud.

Programming Languages

PythonJavaJavaScriptTypeScriptSQL

AI & Machine Learning

Machine LearningDeep LearningComputer VisionNLPLLMsRAGAgentic AIFederated LearningMulti-Agent SystemsKnowledge Graphs

Frameworks & Libraries

PyTorchTensorFlowScikit-LearnLangChainLangGraphFastAPIReactNext.jsNode.js

Databases & Vector Stores

PostgreSQLMongoDBSQLiteSupabaseChromaDB

MLOps & Tools

GitGitHubDockerLinuxOllamaLangfusedeepevalREST APIs

Core CS Concepts

Data Structures & AlgorithmsOOPDBMSOperating SystemsComputer Networks

Flagship Projects

Production GenAI systems, evaluated like production systems

Both grounded in UAE regulatory context — Dubai tenancy law and DFSA/DIFC registers — and both shipped with an evaluation harness, not just a demo.

Built for UAE — Dubai Tenancy Law

Sakan

Grounded RAG Assistant for Dubai Tenancy Law

Built a hybrid retrieval-augmented generation system that answers Dubai tenancy law questions with citations to the exact legal article, merging vector search (ChromaDB) and keyword search (BM25) via Reciprocal Rank Fusion over law text chunked by article boundary.
Resolved a legal-correctness edge case where an amendment law superseded specific articles by tagging every chunk with instrument, article number, and currency status, excluding superseded text from the retrieval space to prevent outdated-article citations.
Built a citation-grounding verifier that checks every cited article against retrieved source text, catching hallucinated citations before they reach the user.
Deployed publicly on Streamlit Cloud using Claude Haiku for cost-efficient, context-grounded generation.

PythonChromaDBBM25Anthropic APIStreamlitdeepeval

Live Demo GitHub

0%Refusal Accuracy

0%Citation Grounding

0.00Avg. Faithfulness Score

Built for UAE — DFSA / DIFC Registers

Scout

Autonomous Multi-Agent Company Due-Diligence Tool

Designed a 5-node LangGraph state machine (plan → act → cluster → synthesize → critique) that researches a company across web/news search and regulatory registers (DFSA, DIFC), producing a structured, Pydantic-typed due-diligence report.
Implemented a deterministic verification layer requiring 2+ independent source domains to mark a claim “verified” vs. “flagged,” deliberately decoupled from LLM judgment to keep the trust decision auditable.
Instrumented full request tracing with Langfuse (per-node spans, cost/latency capture) and built retry/backoff handling for flaky search calls and malformed structured outputs.
Deployed as a FastAPI backend + Streamlit UI across two services on Render.

PythonLangGraphPydanticFastAPITavily APILangfuseClaude Sonnet

Live Demo GitHub

0%Tool-Call Success Rate

0.0%Factual Recall

~0sAvg. Latency

~$0.00LLM Cost / Report

Other Projects

Before the UAE-market focus

Computer vision, knowledge graphs, and privacy-preserving ML — the range this specialization was built on.

DriveLegal

Graph-Powered Indian Road Law Assistant

AI-powered legal assistant built on a structured knowledge graph of traffic violations, penalties, vehicle categories, and Motor Vehicle Act references.
Multilingual AI assistance and automated fine-calculation workflows.
Interactive 3D graph visualization for exploring legal relationships.

FastAPIExpoSQLiteKnowledge Graphs

GitHub

AI-Enabled CCTV Person Tracking

Real-Time Multi-Camera Surveillance System

Real-time person detection and tracking system across CCTV video streams using computer vision and deep learning pipelines.
Robust under occlusion, viewpoint variation, and challenging lighting conditions.
End-to-end monitoring framework for multi-camera movement analysis.

PythonComputer VisionDeep Learning

GitHub

Federated Learning for Breast Cancer Prediction

Privacy-Preserving Collaborative Model Training

Privacy-preserving federated learning framework enabling collaborative model training across decentralized healthcare datasets without sharing patient data.
Optimization techniques applied to improve model convergence and communication efficiency.
Manuscript under preparation.

PythonPyTorchFederated Learning

GitHub

What's Next

Gaps I found, not gaps I'm hiding

Real limitations Sakan and Scout's own evaluation harnesses surfaced — documented in each project's README, not shipped yet.

Scout

Cross-Cluster Contradiction Detection

Scout's verifier currently checks each claim-cluster in isolation — 2+ independent domains marks a claim "verified." The eval surfaced a real gap: when two sources disagree on a number (e.g. two different reported acquisition prices), each value can independently clear the bar and get marked verified without the contradiction ever being flagged. Fixing this means detecting when two clusters assert different values for the same underlying fact, not a one-line patch — the most concrete next build.

Sakan

Multi-Emirate Legal Coverage

Sakan is deliberately scoped to Dubai tenancy law (Law 26/2007, Law 33/2008, Decree 43/2013) and explicitly refuses out-of-scope questions — ask it about Abu Dhabi rent law today and it correctly says so isn't covered. Extending the same clause-tagged, hybrid-retrieval architecture to Abu Dhabi and Sharjah tenancy law is the natural next corpus to ingest.

Sakan

Arabic Bilingual Support

Flagged as a stretch goal during Sakan's original scoping and deliberately cut to hit the build timeline. Real UAE tenancy disputes happen in both English and Arabic, so a bilingual retrieval + citation path — not just a translated UI — is the honest next step for the project to be genuinely useful, not just demo-complete.

Experience

Where the hands-on work happened

AI/ML Intern · InTrain Tech

Dec 2024 — Feb 2025

Developed an AI-powered gunny bag detection system for rice warehouses using YOLO-based object detection models.
Created and annotated custom datasets of rice gunny bags and applied data augmentation techniques to improve model generalization.
Trained, evaluated, and fine-tuned deep learning models to achieve robust detection performance across varying warehouse environments.
Designed a real-time detection pipeline for automated inventory estimation and warehouse monitoring, reducing manual inspection and counting effort.

Research

Published work

SiRouteNet: Learning Signal-Integrity-Aware Cost Maps for Constraint-Driven PCB Autorouting

Co-authored an AI-assisted PCB autorouting framework combining a U-Net-based signal-integrity prediction model with deterministic A* routing.
Designed and evaluated deep learning models for signal-integrity risk estimation in PCB layouts, demonstrating reduced routing risk while preserving geometric efficiency.

Achievements

Hackathon results

1st Place

Smart Automation, NRIIT National Hackathon 2024

3rd Place

Hack-to-Detect AI Surveillance System Hackathon

Top 10 Finalist

International VIT-AP HackByte Hackathon

Finalist

BIMSTEC AI Road Safety Hackathon 2026 — Centre of Excellence for Road Safety (CoRES), IIT Madras, 16 Jul 2026

Education

Academic background

Vellore Institute of Technology, Andhra Pradesh (VIT-AP)

Bachelor of Technology — Computer Science & Engineering (AI & ML Specialization)

Amaravati, India

2022 — 2026CGPA 8.2 / 10

Contact

Let's talk

Open to AI/ML and GenAI Engineer roles in the UAE — happy to walk through Sakan or Scout's evaluation results in more depth.

Dubai, UAE · Immediately Available · Open to Relocate within UAE

Emailsantoshmuruggu@gmail.com Phone+971 55 340 5571 GitHubgithub.com/santoshmuruggu LinkedInlinkedin.com/in/santoshmuruggu Instagraminstagram.com/miles0_07