01Build
Backend systems
Services with clean contracts, observable behavior, and room to grow.
PythonFastAPIREST
professional profile / selected work
Software Engineer
Umar Balak
I'm a backend-focused software engineer building calm systems at the intersection of AI, infrastructure, and real-world complexity.
the short version
Less noise. Better abstractions. Software that holds up after the demo.
01
based in
Hyderabad, India
02
currently at
Cognizant
how I work
A small, sharp toolkit.
The tools change. The principles stay: make it useful, make it legible, make it hold.
02Ship
Cloud infrastructure
Practical platforms that make reliable delivery feel routine.
GCPAzure
03Explore
AI & data
Useful intelligence grounded in good data flows and thoughtful evaluation.
RAGVertex AISQLVector search
04Care
Engineering practice
A bias toward simple decisions, clear docs, and software that lasts.
System designObservabilitySecurity
selected work
Built for the messy middle.
01
work / ai operations
AIRA
An AI-driven ticket-triage system using Gemini to score ServiceNow incident completeness and generate targeted clarification questions.
A production-ready RAG workflow grounded in similar incidents, with parallel batch processing, an on-demand API, and auditable failure isolation.
GeminiRAGBigQueryServiceNow
02
work / data operations
Control-M
A production BigQuery pipeline tracking the complete daily change history of enterprise batch jobs.
A single comparison layer classifies job changes while protecting downstream data from ambiguous keys, duplicates, and alert noise.
BigQuerySQLData quality
03
open source / deep learning
VanillaNets
A zero-dependency neural network library built from scratch in pure Python and NumPy.
An inspectable Keras-style API with explicit gradients, fused softmax loss, modern optimizers, and a documented PyPI release.
PythonNumPyDeep learningPyPI
04
exploration / knowledge systems
Cognify
A collaborative research workspace for building peer-reviewed knowledge bases with RAG.
Research tools with memory, context, and receipts.
FastAPILangChainPinecone
05
exploration / privacy ml
AdaptFL
Federated learning for training models without moving sensitive data out of its source.
Privacy as a design constraint, not an afterthought.
TensorFlowDifferential privacyFastAPI
a little context
Curious by default.
I started with machine learning because I wanted to understand what happens beneath the surface. That curiosity eventually pulled me toward backend engineering, distributed systems, and cloud infrastructure.
Today, I like working where product ideas meet operational reality: designing clear data flows, resilient services, and tools that make complicated work feel ordinary.
ai systemsdata platformscloud native