Lights out and away we go
P1Lights Out
Software Development Engineer · Backend & AI Infrastructure
I build RAG agents and backend systems that run in production — and contribute to the open-source tools behind them.
Software Development Engineer @ Quadrant Technologies (client: Microsoft) · Open to software engineering roles
70+
Production workflows owned
100K+
API requests / day — Doppular
<200ms
Vector search latency
1M+
Telemetry records / day
BestPersonal bestStandard
P2Driver Profile
Fast systems. Built to finish.
Backend and AI infrastructure is my home circuit. At Quadrant Technologies, working for Microsoft, I build RAG agents on Azure AI Foundry with LangGraph and MCP, and own a Python / Airflow framework that keeps 70+ production workflows running.
Before that I co-founded Doppular — a social app that matches you with people around the world who look like you — and built its distributed backend to handle 100K+ API requests a day, with vector search answering in under 200ms.
My style is root-cause over patches: reproduce it first, fix where it actually breaks, and ship the regression test with the fix. Same discipline as a race weekend — preparation wins, guesswork loses.
Avinash Kalamati
Software Development Engineer
P3Race Calendar
Four drives, real numbers. Every result below shipped to production or publication.
A social app that matches you with people around the world who have similar faces — “alike in looks, unique in life.”
Round 1 / 4
P4Car Setup
Read it like a car setup sheet — what's under the bodywork, grouped the way I'd actually reach for it on a build.
Power Unit
Chassis
Aero
Electronics
Highlighted = core stack I use most
P5Podium
Open-source results — called exactly as they stand. None of these are merged yet; the approvals, green CI, and downstream validation below are the real scoreboard.
P1 Flagship build
testcontainers-python · PR #1145
A brand-new fake-gcs-server container module for the testcontainers ecosystem — my first open-source PR. Approved by a maintainer and validated downstream in the panodata/omniload suite, where all 3 GCS integration cases passed unchanged as a drop-in replacement.
P2 All green
Azure-Samples/azure-search-openai-demo · PR #3250
On the free Azure AI Search tier, results without a reranker score were treated as 0 and filtered out — so search returned nothing. This fix keeps those results. Microsoft CLA signed.
P3 Root-cause fix
run-llama/llama-agents · PR #838
“SwEu” was deserializing to 1 — deserialization tried pickle before JSON and base64 decoding lacked validation. My fix addresses the root cause, not just the symptom, and ships with regression tests. The most complete of three competing fixes.
WIP On the surrounds
Claimed and in progress: on resume, the Functional API re-runs a @task whose exception the entrypoint already caught, instead of replaying its recorded outcome. Checkpoint / task-replay semantics, with a self-contained repro using InMemorySaver.
P6Paddock
Bikes, cars & anything with a throttle
Off the clock I'm an autophile through and through — superbikes, race liveries, carbon everything. Sport is the mindset I bring to code too: train, measure, shave the lap time, repeat.
Current target — first motorcycle · saving in progress
P7Team Radio
“Radio check. I’m listening.”
Open to software engineering roles — backend, AI infrastructure, and platform work. Fastest way to reach me is email or LinkedIn; my code is all on GitHub.
Lap1 / of 7Home
←→ scroll or swipe