AI & Data Engineer

Saatvika
Chokkapu

Building RAG systems and the data pipelines behind them — explainable, not just performant.

RAGLLMPythonDatabricksSQL
rev 2.0 · aug 2026
SCROLL ↓
01 — THE JOURNEY

How I got here

JUN 2022 — NOV 2022 · VISAKHAPATNAM, INDIA

Data Analytics Intern

NSR Industries
Where the data habit started.
  • Analyzed 10,000+ raw-material and production records with SQL and Pandas for weekly operations projections.
  • Built 20+ recurring Power BI dashboards and tracked 10+ KPIs in Tableau across production, purchasing, and sales.
SQLPandasPower BITableau
DEC 2022 — JUL 2024 · VISAKHAPATNAM, INDIA

Associate Data Engineer

NSR Industries
Kept 57+ production pipelines running every morning.
  • Ran 57+ Airflow pipelines across three environments, turning 100K+ raw operational records into dbt-modeled datasets teams trusted daily.
  • Handled 30+ production failures across two on-call rotations — root causes traced through Airflow logs and SQL, not just restarts.
  • Replaced a 48-sheet Excel runbook with a FastAPI + PostgreSQL app adopted by 30+ colleagues.
AirflowdbtSQLPythonPostgreSQLFastAPI
DEC 2025 — MAY 2026 · REMOTE

AI Engineer Intern

Sonus Software Solutions Inc.
Took a RAG Q&A agent from prototype to production.
  • FAISS prototype → production on Pinecone, serving semantic search over 10,000+ documents for engineering and delivery teams.
  • Built the retrieval layer with LangChain, Voyage AI embeddings, and FastAPI — tested chunking and prompting against 75+ real questions to keep answers grounded.
  • Weekly Databricks ETL from Jira and SharePoint kept the knowledge base fresh, with Power BI tracking pipeline health for a 20-person team.
RAGLangChainPineconeFAISSFastAPIDatabricksPower BI
02 — SELECTED WORK

Projects

Three systems, three different problems — still very much in motion.

FlowMind — network intrusion detection
Work in progress · rev 2.0

FlowMind

Intrusion detection that explains itself — every flagged threat grounded in MITRE ATT&CK evidence.

2.8M+ flows · 94.1% F1
view project ↗
Verdict — cost-sensitive fraud decisioning
Work in progress · rev 2.0

Verdict

Fraud decisions priced in dollars — approve, decline, or review by expected cost, in real time.

590K+ transactions · p50 ~28ms
view project ↗
Artistry vs. Popularity — music analytics
Work in progress · expanding

Artistry vs. Popularity

25 years of Billboard, Grammy & Spotify data asking one question: has streaming made music disposable?

−44% avg song duration · 66% one-hit wonders
view project ↗
03 — CREDENTIAL

Certification

NEW · AUG 2026
db

Databricks Certified
Data Engineer Associate

ID 192184056 · VALID THROUGH AUG 2028 · VIEW CREDENTIAL
↗
DatabricksApache SparkDelta LakeUnity CatalogETL
04 — WHAT I’M LOOKING FOR

Roles & the skills behind them

R—01

Data Engineer

Pipelines that run every morning without anyone thinking about them.

PythonSQLSparkAirflowdbtKafkaSnowflakeTerraform
Receipts57+ production Airflow DAGs at NSR · Kafka → Iceberg lakehouse in Verdict · Databricks certified.
R—02

AI Engineer

Retrieval systems that answer with evidence you can check.

RAGLangChainPineconeQdrantLLM APIsFastAPIEvaluation
ReceiptsRAG agent to production at Sonus (10K+ docs) · MITRE-grounded explanations in FlowMind · 75+ question eval sets.
R—03

Data Analyst

Numbers that answer questions people are actually asking.

SQLPythonPandasPower BITableauStatistics
Receipts20+ recurring Power BI dashboards at NSR · 10+ KPIs in Tableau · 25-year music industry analysis.
R—04

Business Analyst

Translating between what teams need and what data can say.

SQLExcelPower BIRequirementsStakeholdersProcess
ReceiptsReplaced a 48-sheet Excel runbook with an app 30+ colleagues adopted · weekly ops projections leadership planned around.
05 — OFF THE CLOCK

When I’m not shipping pipelines

sketchingdesigningmuseumsreadingtennisvolleyball

The other half of the week, in no particular order.

it never loses. neither do i, eventually.