// OPERATOR
Rohan Keshav
Software Engineer
SOFTWARE + DATA PIPELINES
CLOUD-NATIVE ON AWS
AGENTIC AI WORKFLOWS
ROHAN
"Ship the messy first pass.
Let the iterations do the heavy lifting."
// SYSTEM_PROFILE.MD
>
cat system_profile.md
Software engineer who builds reliable systems. Strong on backend services, data flow, and the unsexy plumbing that keeps things alive at 3am. I care less about the trend cycle and more about the parts that quietly hold a product together.
Daily driver is Python for backend services, with SQL and event-driven pipelines holding the data layer together. Years spent profiling slow paths, refactoring legacy modules into clean abstractions, working with APIs that don’t leak complexity, and shipping the kind of changes that show up on a graph the next morning.
Increasingly drawn to AWS as the deployment target of choice and agentic AI as a way for small teams to ship big work. Currently exploring how S3, Glue, Redshift, and Fargate slot into modern data flows, and where LLM-driven agents earn their keep beyond the demo phase. Always looking for the spot where solid engineering meets the new stack.
Currently
Shipping backend systems
Philosophy
Idea → Build → iterate → Repeat
Exploring
AWS and agentic AI
■ EDITORIAL / ACADEMIC_RECORD
KNOWLEDGE
BASE
An editorial spread on formal CS formation.
Academic dispatch
Formal CS training meant to be read, not skimmed for buzzwords.
UTA
Aug 2023 — May 2025 · Arlington, TX
University of Texas at Arlington
Master of Science, Computer Science
Tracks in Intelligent Systems, Cloud Computing & Databases—coursework toward reliable ML-backed systems and data-heavy backends.
Cover story · Latest degree
VIT
Aug 2018 — Jul 2022 · Bangalore, India
Vemana Institute of Technology
Bachelor of Engineering, Information Science
Concentration in Artificial Intelligence & Machine Learning—foundational theory and practice before graduate depth.
■ SYSTEM.LOG / CAPABILITY_MATRIX
CAPABILITY
MATRIX
Hover a tile to light the logo and reveal the skill—idle tiles stay muted for a calmer scan.
Capability matrix · brand marks · hover to identify
Languages
Data + Cloud
ML / DL
AI & Automation
UI / Frontend
Tools & DevOps
■ SYSTEM.LOG / OPERATIONS_LOG
OPERATIONS
LOG
Reverse-chronological roles—role, org, tenure, stack, and impact lines recruiters scan first (ownership + metrics, CAR-style bullets).
[ ORDER: NEWEST FIRST ]
[ STRUCTURE: ROLE · IMPACT · STACK ]
[ SIGNALS: METRICS + OWNERSHIP ]
Software Engineer
Arcadis — City of Houston
- Refactored legacy backend into a parallel compute pipeline with Python Multiprocessing, reducing execution time by ~33% while holding memory footprint steady.
- Automated cloud-native data infrastructure on AWS (S3, Glue, Redshift), turning raw data into analytics-ready datasets and removing manual processing bottlenecks.
- Streamlined backend data pipelines on AWS Fargate, processing 2M records daily and cutting pipeline execution time by 18%.
AI Engineer
University of Texas at Arlington
- Designed agentic AI workflows using Python and LLMs, automating decision-making pipelines and reducing manual research iteration cycles by 1.5x.
- Integrated AI agents with digital signal processing, building adaptive systems that dynamically improved contextual signal interpretation.
Graduate Teaching Assistant
University of Texas at Arlington
- As Cloud Computing and Operating Systems enrollment scaled, students needed reliable mentorship beyond lecture, my role was to own recurring touchpoints where 200+ learners could get clarity on architectures, concurrency, memory/process models, and lab debugging without lowering rigor.
- Ran weekly Q&A sessions that surfaced recurring misconceptions early; reframed questions into teachable mini-lessons, coached toward independent troubleshooting, and fed themes back to faculty so pacing and assignments stayed aligned with where the cohort struggled.
- Partnered on course delivery by assisting with exam proctoring and assessment logistics, maintaining integrity and consistent instructions under pressure so students knew exactly how to seek legitimate help during high-stakes evaluations.
Data Science and Business Analytics Intern
The Sparks Foundation
- Architected a supervised ML prediction system, evaluating Linear Regression and clustering models through comparative analysis and reaching 89% prediction accuracy on student exam outcomes from preparation signals—a data-driven baseline for academic planning.
- Built end-to-end MySQL ETL pipelines that fed clean, structured data into the model and drove a 16% precision improvement through systematic feature engineering and EDA.
Software Developer Intern
Campus Build
- Refactored production C++ modules applying OOP, Functional Programming, and STL principles to improve modularity and maintainability.
- Contributed to full-stack feature development in an agile startup environment, joining code reviews and SDLC ceremonies alongside senior developers.
■ SYSTEM.LOG / EXECUTION_HISTORY
EXECUTION
HISTORY
Spotlight on five flagship builds—full briefs below. Five more ships listed as honorary mentions.
[ FEATURED: 05 ]
[ HONORARY: 05 ]
[ TOTAL BUILDS: 10 ]
AGENTIC RESUME OPTIMIZER
ReAct AGENT · RAG · ATS-OPTIMIZED OUTPUT
Autonomous AI platform with a ReAct agent loop and 14 OpenAI function-calling tools for end-to-end JD analysis, gap detection, and ATS-optimized resume output across sessions—translating a complex research idea into a production-grade codebase. Per-user RAG knowledge base with SQLite vector storage and OpenAI embeddings for high JD alignment; template-preserving LaTeX pipeline with brace-aware structural repair and automated one-page enforcement.
Python
OpenAI API
ReAct
RAG
SQLite
LaTeX
TELEGRAM AGENT
AUTONOMOUS SCHEDULING · n8n × LLM
Telegram-integrated automation workflow in n8n using Python, LangChain, JSON, and OpenAI APIs — streamlining natural-language scheduling with Airtable context. Built a voice-to-text pipeline in JavaScript with LLM intent parsing, letting AI agents autonomously create and manage Google Calendar events.
n8n
LangChain
OpenAI
JavaScript
Airtable
QUANT VISION
REAL-TIME ANALYTICS · BACKTESTING ENGINE
End-to-end fintech dashboard in Python with Streamlit and Plotly delivering real-time candlestick visualizations and analytics for equities and derivatives. Added backtesting and time-series forecasting with Pandas, NumPy, and statistical modeling to surface risk-management insights and improve investment timing.
Python
Streamlit
Plotly
yfinance
Pandas
gRPC COMPUTE SERVER
SYNC + ASYNC RPC · MULTI-THREADED
Multi-threaded gRPC server in Python supporting distributed file operations and compute through sync and async RPCs. A folder-watcher client detects file changes and triggers remote compute, eliminating manual monitoring overhead and enabling real-time workflows.
Python
gRPC
Threading
Distributed
MEGACOSM
0%
perf lift over baseline · IRJCS published
Computer-vision pipeline in Python with YOLO v5 and OpenCV for real-time astronomical object detection and classification. Iterative CNN optimizations and hyperparameter tuning lifted model performance 25% over baseline. Published in IRJCS.
Python
YOLO v5
CNN
OpenCV
NumPy
OPEN A CHANNEL
Cool projects, spicy ideas, or just want to nerd out about agents and pipelines? Drop a line.