Open to AI engineering roles

Pavani Rajula.

AI Engineer & Data Scientist.

specializing in

I design and deploy production AI systems — focusing on agentic workflows, RAG architectures, and evaluation frameworks that ensure reliability, compliance, and hallucination mitigation in client environments.

View my work Download resume (PDF)
Portrait of Pavani Rajula

01 · about

Applied AI engineering, built to ship.

I specialize in building reliable, verifiable AI applications: production RAG services, agentic workflows, and domain-grounded architectures with robust evaluation harnesses and compliance guardrails.

Through my company, NeuCorelytix Solutions, I work directly with engineering teams to design and ship AI solutions — most recently as an external consultant to Data Migration International.

Previously, I spent two years developing production LLM/RAG systems and GDPR-compliant data workflows at Data Migration International in Switzerland, and three years engineering regulatory data pipelines over 20 TB of Hadoop data at State Street.

My background is rooted in open source. As an Outreachy intern at Ceph, I built the CephFS Shell (still maintained in official docs), and have presented technical work at PyConf, DevConf, and PyTorch Conference Europe. Off the keyboard: singing, yoga, badminton, and art.

02 · skills

Technical stack & core competencies.

From agentic architectures and retrieval pipelines to backend microservices, data infra, and cloud deployment.

LLM & Agentic Engineeringai-llm/
Agentic AIRAGMCP Tool OrchestrationPrompt Engineering LangChainLlamaIndexAgno OpenAI APIHF TransformersSemantic Search
Evaluation & Reliabilityevals/
LLM Eval MetricsHallucination Mitigation MonitoringFeedback LoopsMLflow Inference Optimization
Client Delivery & Solutionsfield/
Client-Embedded ConsultingRequirements → Production Rapid PrototypingEnd-to-End Ownership GDPR & Compliance DomainsDemos & Technical Talks Cross-Functional Delivery
ML / Deep Learningml-dl/
PyTorchTensorFlowscikit-learn Computer VisionNLPReinforcement Learning Explainable AI
Backend & APIsbackend/
PythonFastAPINestJS REST APIsMicroservicesAsync Pipelines CeleryRedisJavaSQL
Data & Retrieval Infradata/
QdrantVector SearchEmbeddings MongoDBMySQLHive HadoopSpark
Cloud & MLOpsinfra/
DockerAWSS3 · Lambda · IAM AzureDistributed Task ExecutionUNIX / Shell
Frontend & Prototypingui/
Next.jsReactStreamlit Node.jsTurborepo

03 · selected work

Selected projects & system architectures.

Problem, architecture, and technical outcomes across production AI and research systems.

PAV·AI · the RAG chatbot on this page

live · RAG
ProblemRecruiters want quick answers about me without scrolling. And I wanted to demonstrate a grounded RAG, not just list it as a skill.
ApproachA full RAG over my own portfolio: ingest this site, embed, retrieve, grounded generation on a free Cloudflare Worker, with prompt-injection guardrails and an eval harness.
OutcomeThe "Ask PAV·AI" button here. Retrieval you can see, refuses when it doesn't know, tested for faithfulness.

Cloudflare Workers AI · bge-small · Llama-3.1-8b-fp8 · cosine retrieval · evals · $0/mo

read the case study →try it ↘

VakilSarathi · Legal Intelligence Engine for India

legal AI · bi-temporal RAG · open-source
OverviewA foundational, open-source legal intelligence engine modeling Indian statutory law (GST initially, expanding across Income Tax, Companies Act, FEMA) as structured, canonical, versioned knowledge objects for verifiable AI research and drafting.
ArchitectureThree-Plane Design: Knowledge Plane (Docling/OCR ingestion, PostgreSQL system of record with bi-temporal range types, Neo4j knowledge graph, and Qdrant + OpenSearch hybrid retrieval), Matter Plane (tenant-scoped with Presidio/GLiNER PII firewall & vault), and Model Plane (stateless open-weight reasoning models like Qwen & DeepSeek via vLLM).
Key InnovationBi-temporal correctness (point-in-time statutory versioning) and Legal Object URIs (LOUs) ensuring hallucination-proof legal citations, automated SCN replies, and 100% self-hostable air-gapped deployment.

Python 3.12 · FastAPI · PostgreSQL 16 · Neo4j · Qdrant · OpenSearch · BGE-M3 · Docling · Presidio · vLLM · Temporal · MinIO · Docker

Re:Me · AI Browser Agent & Personal Memory

agentic · browser agent · RAG
OverviewAn AI-powered browser extension that remembers useful things encountered online, understands their context, and surfaces them proactively when they become relevant.
ArchitectureCombines an autonomous browser agent with a personal memory layer: agentic workflows with multimodal content understanding, semantic retrieval over vector embeddings, and knowledge graph entity relationships for context-aware recommendations and web automation.
HighlightA browser agent that understands what you're currently looking for and transforms passive web bookmarks into actionable reminders, connected knowledge, and timely resurfacing.

Browser Extension · LLMs / GenAI · Agentic Workflows · RAG · Vector DB · Knowledge Graphs · Web Automation · Multimodal AI

Memoiary · Multimodal AI Journal & Persistent Memory

multimodal AI · long-term memory · full-stack
OverviewA multimodal AI journal designed as a persistent personal memory system. Captures text, voice, images, and daily moments to let users revisit life experiences with rich semantic context.
ArchitectureEnd-to-end multimodal ingestion pipeline featuring speech-to-text, automated entity extraction & resolution (people, places, events, emotions, concepts), and knowledge graph relationship modeling across time with contextual retrieval over vector embeddings.
HighlightA persistent multimodal memory layer that connects what you write, say, see, and experience over time, backed by a responsive Next.js and Firebase architecture.

Next.js · Firebase / Firestore · LLMs / GenAI · Multimodal AI · Speech-to-Text · RAG · Knowledge Graphs · Vector Search

code & demo on GitHub →

PIP-Net Fusion · Ensembling & Model Federation

MSc thesis
ProblemEnsembles boost accuracy but usually destroy interpretability. That's a dealbreaker in high-stakes ML.
ApproachFused multiple interpretable PIP-Net models via ensembling and federation strategies, preserving prototype-based transparency.
OutcomeHigher accuracy without sacrificing explainability. Extended as a poster at PyTorch Conference Europe 2026.

PyTorch · XAI · Federated Learning · Computer Vision

code on GitHub →poster: PyTorch Conf EU 2026

Decoding Word Embeddings · Interpretability Toolkit

NLP research
ProblemEmbeddings are powerful but opaque. Hard to see what their dimensions encode.
ApproachExtended SensePOLAR to support multiple oracles, models, and languages, with a Streamlit interface for rapid experiments.
OutcomeA reusable tool researchers use to compare interpretability across configurations.

Hugging Face · Streamlit · Word Embeddings · Python

code on GitHub →

04 · experience

The whole journey, one timeline.

Work, study, and open source on one timeline. Teal dots are roles, gold dots are degrees.

NeuCorelytix Solutions

Founder · AI DeveloperworkSep 2025 · present

Remote · my own company. External consultant to Data Migration International until Jul 2026

  • Building full-stack AI applications: RAG pipelines, conversational agents, and LLM-integrated web platforms with frontend (Next.js/React) and backend (NestJS/FastAPI) services.
  • Architecting AI microservices: Python/FastAPI endpoints for retrieval-augmented generation, decoupled from transactional backends; CMS-driven knowledge bases with automated vector re-indexing.
  • Engineering scalable monorepo architectures with Docker Compose orchestration across multiple interconnected services.
  • Driving AI integration at client environments: GDPR-compliant PII identification, data-quality systems, and production ML pipelines (under NeuCorelytix Solutions).

Data Migration International

Data Science & AI Developer · external consultant via NeuCorelytix SolutionsfreelanceSep 2025 · Jul 2026

Hyderabad, India · on-site

  • Develop agentic AI systems with LLMs, RAG, tool orchestration — enterprise system chatbot understanding data, view relations.
  • Design 3+ scalable AI service architectures: microservice model endpoints, async pipelines, distributed task execution.
  • Build evaluation frameworks covering 10+ agent behaviors and failure scenarios.
  • Implement monitoring and feedback loops — reduced hallucination rate by 60% across client systems.
  • Integrate AI into client environments ensuring compliance, security, and cost efficiency.

Data Migration International

Data Science & AI DeveloperworkOct 2023 · Sep 2025

Kreuzlingen, Switzerland · hybrid

  • Developed and deployed 4+ AI solutions with LLMs, RAG, NLP, and structured data pipelines.
  • Fine-tuned and optimized 5+ models, improving inference speed by 40%.
  • Built end-to-end ML pipelines.
  • Contributed to model deployment across client environments.

University of Mannheim

MSc, Data ScienceeducationFeb 2022 · Jun 2024

Mannheim, Germany · completed alongside full-time AI work

  • Thesis: PIP-Net Fusion on interpretable model ensembling — 3–5% accuracy gain over single models while preserving explainability; extended as a poster at PyTorch Conference Europe 2026.
  • Coursework: Deep Learning, Computer Vision, Reinforcement Learning, Information Retrieval, Text Analytics, Network Science.

State Street Corporation

Associate 2 · Data EngineerworkJul 2019 · Feb 2022

Hyderabad, India

  • Owned six regulatory reports on Axiom Controller View across the full SDLC.
  • Automated reporting with SQL/UNIX/shell for 40% less manual workload.
  • Python over 20 TB Hadoop data for 50% faster processing; AWS migration PoC for 30% lower infra cost.

BVRIT Hyderabad College of Engineering for Women

B.Tech, Information TechnologyeducationAug 2015 · Jun 2019

Hyderabad, India

  • Algorithms, Machine Learning, Big Data Analytics, Cloud Computing, Databases, Software Engineering.

Ceph Organization

Outreachy Intern · Open Sourceopen sourceMay · Aug 2018

Remote

  • Built the CephFS Shell: Unix-style commands over libcephfs Python bindings — still maintained in official Ceph docs 8+ years later. See it here →

05 · writing

Technical notes & articles.

Notes on agentic architectures, production RAG, evaluation frameworks, and real-world system design lessons.

✍ just getting started

The first essays are on the way. Subscribe to catch them the moment they drop. This list fills in automatically as I publish.

Visit the Substack →

06 · recognition

Recognition along the way.

Winner · JIVS Hackathon: All About Data & AI

Data Migration International

Geo-Masking privacy system preventing identification of people and locations in shared images.

Published · CLEF 2022 Working Notes

CLEF Conference

"Stacked Model based Argument Extraction and Stance Detection using Embedded LSTM" with a conference talk. read the paper →

Poster · PyTorch Conference Europe 2026

PyTorch Foundation

"When Models Collaborate but Data Cannot: Explainable Ensemble Learning Under Privacy Constraints."

3× Linux Foundation Scholarships

OSS Summit EU 2022 · KubeCon EU 2020 & 2022

Recognized for sustained open-source and cloud-native community contributions.

Speaker · DevConf.in & PyConf Hyderabad

2019

Ceph at DevConf.in Bengaluru; "Open Source for Beginners" at PyConf Hyderabad.

Outreachy 2018 Participant

Software Freedom Conservancy

Selected for the competitive open-source internship; contributed core CLI tooling to Ceph. see it in Ceph docs →

07 · contact

Let's build something intelligent together.

Agentic systems, RAG at scale, or an AI product that needs to ship. My inbox is open.

Say hello Download resume (PDF) Connect on LinkedIn
emailrajulapavani@outlook.com in/rajula-pavani gh/rpavani1998 substack/pavanirajula