APR 2026
— AUG 2026
Software Developer I, AI Solutions
Kinaxis · Toronto, ON
Built multi-agent orchestration, secure MCP tools, and an NL-to-MQL workflow for supply-chain queries.
01AI ENGINEER · SYSTEMS THINKER
I build AI for
the messy middle.
When a good idea meets real work—tools, context, handoffs, edge cases—I build the system that gets it through.
FIELD STUDY 001FROM INTENT TO SOFTWARE
01 / INPUT
Intenthuman request
02 / REASON
↗↘·
Agent loopplan · act · check03 / USE
+
Real toolscontext · action04 / VERIFY
✓
Useful changereviewable output↳HUMAN CHECKPOINT
THE PROMPT IS THE STARTING POINT.NOT THE PRODUCT.
FIG. 01
00 CURRENT CHAPTERWHERE I’M BUILDING NOW
KINAXIS · TORONTO
Software Developer I,
AI Innovation
AUG 2026 — PRESENT
Building the platform behind agentic AI for supply-chain work.
I’m developing a production-grade system that takes a task through Intake → Design → Build → Review, with human approval gates built into the workflow. The work spans cloud-ready deployment, secure tool use, and observability.
01Intake
→02Design
→03Build
→04Review
KUBERNETES / AZURE AKS / GOOGLE GKEDATABRICKS / A2A + MCP / OPENTELEMETRY
01 EXPERIENCE / 2022—NOW
From research
to production.
A continuous thread: make intelligent systems useful outside the lab.
~4
YEARS COMBINEDAI · SOFTWARE · RESEARCHIncludes internships and academic research appointments.
SEP 2025
— APR 2026
Intern Developer, AI Solutions
Kinaxis · Toronto, ON
Improved natural-language-to-MQL accuracy to 84% with Gemini 2.5 Pro; built agentic workflows for Kinaxis Maestro.
02SEP 2023
— AUG 2025
Graduate Research & Teaching Assistant
Western University · London, ON
Researched gaze-aware lane-change prediction with object detection and LSTM, GRU, and TCN sequence models.
03JAN 2023
— JUL 2023
Artificial Intelligence Research Assistant
IGDM Lab, University of Calgary · Remote
Implemented graph-based traffic forecasting and an interactive visualization of road-network patterns.
04JUL 2022
— SEP 2022
Globalink Research Intern
MITACS / IGDM Lab, University of Calgary
Applied forecasting models to parcel demand, driver behavior, and speed for regional delivery-network optimization.
05CURRENT ROLE: SOFTWARE DEVELOPER I / AI INNOVATION, KINAXISFULL CAREER DETAILS ON LINKEDIN ↗
02 THE WORKING SET
Built around
real problems.
Four different questions.
Four very different systems.
SELECT A FIELD NOTE 01—04
USE THE ARROW KEYS
OR PICK A PROJECT
DEVX / AGENTIC FRAMEWORK IN THE LOOP
WORK ITEM / 08:42
DEVELOPER INTENT
“Add a validation step before
the workflow creates a change.”
01Intakeunderstand
→02Planbreak it down
→03Buildmake the change
→04Reviewhuman checkpoint
✓ CONTEXT GATHERED↗ PLAN READY FOR REVIEW▌
INTENT → WORKING CHANGEPROCESS MAP / 01
−30%
CASE 01 / DEVELOPER TOOLSKINAXIS
KXS DevX
Shorten the distance between “I need this” and a working change.
An agentic framework that moves a software task through intake, planning, implementation, and review. Designed around the actual work developers do—not a chat window bolted onto it.
MY FOCUSAgentic workflow design
MEASURED IMPACT30% shorter development lifecycle
HUMAN REVIEW STAYS IN THE LOOP01 / 04
MAESTRO / KNOWLEDGE RETRIEVAL IN DEVELOPMENT
PLANNING / DATASOURCE 02
MAESTRO / GUIDESOURCE 01
PARENT DOCUMENT
RETRIEVE
RAG
CONTEXTGENERATE
RESPONSE
“Here’s the plan change…”
↳Source: Planning Guide, p. 12
GROUNDED IN THE SOURCE
SOURCE → CONTEXT → ANSWERRETRIEVAL MAP / 02
CITETHE SOURCE
CASE 02 / ENTERPRISE AIIN DEVELOPMENT
Kinara RAG
An answer is only useful when you can trace where it came from.
A RAG chatbot for Kinaxis Maestro. Parent-document retrieval keeps generated responses tied to the larger source—not an orphaned fragment—so people can follow the context behind an answer.
CURRENT QUESTIONHow should retrieval preserve context?
IN PROGRESSRetrieval + agentic lifecycle
BUILT FOR SOURCE-GROUNDED ANSWERS02 / 04
JIRAXIS / TASK COORDINATION BUILT
J✳
ORCHESTRATORGemini 2.5 Pro⌕Triageread the issue
↗Plannerroute the task
⌘Tool agenttake action
◷Statekeep the thread
ONE REQUEST / SPECIALIST AGENTSAGENT MAP / 03
ASK→ ROUTE → ACT
CASE 03 / PERSONAL BUILDGOOGLE CLOUD
JIRAXIS AI
A Jira request can need more than one kind of thinking.
A Gemini-powered assistant that hands work between specialist agents. Built with Google Cloud ADK, explicit workflow state, and tool calling so a request can move from issue to action.
SYSTEM DESIGNSpecialists + orchestration
BUILT WITHADK · Gemini 2.5 Pro · A2A
STATEFUL, SPECIALIST COORDINATION03 / 04
DRIVER MANEUVER / TEMPORAL VISION MSc RESEARCH
OTTAWA / LONDON DATAFRAME t−04 → t
SUBJECT
GAZE / OBJECT
SLIDING WINDOW / PREDICTION
LSTMGRUTCN→LANE CHANGE?
GAZE + OBJECTS → NEXT MANEUVERRESEARCH PLATE / 04
LOOKTHEN PREDICT
CASE 04 / APPLIED RESEARCHWESTERN UNIVERSITY
Eyes on
the road.
Can attention and surrounding motion reveal what a driver will do next?
My MSc thesis combined gaze signals and object detection with sliding-window sequence models to study driver maneuvers—connecting what a driver looks at with the motion developing around them.
MODELSLSTM · GRU · TCN
VISUAL SIGNALYOLO / signs, lights, objects
COMPUTER VISION / SEQUENCE MODELING04 / 04
THESE ARE STARTING POINTS.LET’S TALK ABOUT THE DETAILS ↗
03 A DESIGN PRINCIPLE
WHAT HAPPENS
AFTER THE DEMO?
A clever model
is a beginning.
The difficult part is making it dependable in someone else’s hands.
AGive it context
Connect the answer to the real information behind it.
BLet it do useful work
Give it tools and a path through the task—not infinite freedom.
CMake it reviewable
Keep a person in control of the important decisions.
04 BEYOND THE SYSTEMS
43° 39′ NTORONTO, ON
Engineer by work.
Curious by nature.
I’m Ravindranath—a software developer in AI at Kinaxis. I studied computer science at Western University, where my MSc research focused on gaze-aware driver-maneuver prediction.
I like the space between a promising idea and the moment it becomes genuinely useful. That’s where I spend most of my time.
05 / THE NEXT QUESTIONINBOX OPEN
Got a good
problem?
I’m always interested in thoughtful conversations about AI, useful software, and research that goes somewhere.
ravisawane9@gmail.com ↗