Ashwani Jha
CVBook a call
Open to senior IC & tech-lead rolesRemote · India / Dubai

Building systems
that have to keep working.

Senior engineer, seven years on systems where downtime is a business event. I led core search infrastructure at Property Finder — the listings backbone behind ~6M properties across the GCC — and spent six years at Amazon across Just Walk Out, Amazon Pay, and AWS CloudFormation. These days I build agent infrastructure in the open.

Book a callDownload CVLinkedInEmail
7+
Years shipping production systems at Amazon and Property Finder
~6M
Listings re-platformed with zero customer-visible downtime
40%
Operating cost cut across the Amazon Go store fleet
10
Engineers led — four mentored through promotion
What I do

Four things I get handed.

Every bullet below maps to a system further down this page — nothing here is a skill I have only read about.

Large-scale data & search platforms

Re-platforming the systems a business runs on, while they keep serving traffic.

  • Migrated ~6M listings onto DynamoDB + OpenSearch with a Kinesis dual-write and zero customer-visible downtime
  • Continuous reconciliation that diffs source-of-truth, alerts on per-segment drift, and re-migrates automatically
  • Partner ingestion at 29k listings/week — heterogeneous schemas normalised at the edge, zero integration regressions
DynamoDBOpenSearchKinesisSQSGlue / PySparkGoJava

Applied AI & agent platforms

Agents that do real operational work, not demos — with the retrieval and evaluation around them.

  • Built a company-wide agent platform any engineer or PM could use to spin up Slack-native bots over internal context
  • On-call root-cause agent that cut median diagnosis from ~3 days to ~1 hour on the Amazon codebase
  • RAG over org SOPs and runbooks on Bedrock + SageMaker — ~30% off onboarding and operational task time
PythonBedrockSageMakerVector searchRAGMCPAnthropic / OpenAI SDKs

Making production boring

Taking a system that pages people at 3am and turning it into one that doesn't.

  • Led the op-ex program that took 200+ open incidents to near-zero in two months
  • Cut in-store operating cost ~40% at Amazon Go by replacing manual root-causing with real-time anomaly detection
  • Alerting tuned to customer-impact SLOs, dashboards over the full ingestion path, on-call runbooks rebuilt
GrafanaCloudWatchSLOsPrometheusRunbooksIncident response

Leading engineers

Owning a roadmap and the people delivering it — hiring, growth, and technical direction.

  • Led a 10-engineer pod across backend, ML, and data engineering at Just Walk Out
  • Mentored four engineers through promotion; ran 20+ hiring loops
  • Authored AWS reference architectures and training adopted by 200+ engineers globally
Roadmap ownershipHiringDesign reviewMentorshipCross-team delivery
Selected work

Systems at real scale.

Work I led in industry. The numbers are what the team measured, not what looked good on a slide.

Property Finder · Property Core

Listings → Search migration

~6M listings · 6× media latency cut · zero customer downtime

Re-platformed onto DynamoDB + OpenSearch with a Kinesis dual-write and zero-downtime cutover.

DynamoDBOpenSearchKinesisGluePySpark
Amazon · Just Walk Out

Real-time anomaly detection

40% op-cost ↓ · 60% engineering bandwidth freed

Real-time camera + sensor fusion across Amazon Go stores; freed 60% of engineering bandwidth.

JavaPythonKinesisLambdaSensor fusion
Property Finder · Property Core

Digital Successor

Company-wide rollout · Slack + code review automation

Company-wide AI platform — every employee can spin up a Slack-native digital twin agent.

PythonLLMsSlack APICode review APIsMulti-agent
Property Finder · Property Core

Operational excellence

200 → 0 incidents in 2 months

200+ open incidents → near-zero across listings + search platforms.

GrafanaCloudWatchSLO alertingRunbooksObservability
Property Finder · Property Core

External listings ingestion (DLD)

29k listings/week · TPS ceiling unblocked · 0 regressions

29k listings/week — TPS ceiling unblocked, zero integration regressions.

SQSLRU cacheDynamoDBOpenSearchSchema normalization
Amazon · Just Walk Out

AI auto-root-cause agent

~3 days → ~1 hour · median root-cause time

On-call agent diagnosing Amazon-codebase bugs — median root-cause time ~3 days → ~1 hour.

PythonAWS BedrockLLMsAgentic RCA
Amazon · Just Walk Out

Amazon Go — real-time CV/sensor platform

10-engineer pod · 4 promotions mentored · 20+ hires interviewed

Real-time CV/sensor system behind Amazon Go stores; led a 10-engineer pod across backend, ML, and data eng.

JavaPythonKinesisLambdaSensor fusion
Amazon · Just Walk Out

Internal RAG bot

~30% ↓ onboarding + operational task time

RAG over org SOPs and internal docs — ~30% reduction in onboarding and operational task time.

AWS BedrockSageMakerVector searchRAG
Amazon · Amazon Pay

Amazon Pay × Alexa voice payments

40% transaction failures cut · Alexa voice payments shipped

40% transaction failures cut on Alexa voice-payments via latency profiling + microservice tightening.

Spring BootEKSLambdaAPI GatewayCDK
Amazon · AWS CloudFormation

AWS CloudFormation — IaC patterns

70% release-cycle time ↓ · 200+ engineers trained globally

Cut EKS release-cycle time 70%; training material adopted by 200+ engineers globally.

CloudFormationCDKECSEKSCodeDeploy
Experience

Where I've done it.

Seven years across two companies that both had to keep serving traffic while we changed the engine.

Apr 2025 — Jul 2026
Senior Software Engineer · Property Core
Property Finder · Dubai

Led the team that owned property publishing and search across ~6M listings in the UAE, Bahrain, Qatar, Saudi Arabia and Egypt. Scope spanned ingestion, indexing, relevance, on-call, and the company-wide AI platform.

2022 — Mar 2025
SDE II · Just Walk Out
Amazon · Bangalore

Owned the real-time CV/sensor platform behind Amazon Go's checkout-free stores. Led a 10-engineer pod; mentored four through promotion; conducted 20+ hiring interviews.

2020 — 2022
SDE · Amazon Pay
Amazon · Bangalore

Shipped Amazon Pay × Alexa voice payments and cut transaction failures by ~40% via latency profiling and microservice tightening.

2019 — 2020
Cloud Infrastructure Engineer · CloudFormation
Amazon Web Services · Bangalore

Global SME for CloudFormation, ECS, EKS, and CodeDeploy. Authored reference architectures and training adopted by 200+ engineers worldwide; cut EKS release-cycle time by ~70%.

Education
B.E. Computer Science
Ramaiah Institute of Technology · Bangalore

The full history, with dates and stack detail, is in the CV.

Download CV ↓
Built in the open

Infrastructure I'm building outside work.

Each of these started as a frustration while building agent products on the job. The same primitives kept reappearing, so I extracted them.

L
Live

Leadgen

Pull prospects from many sources, enrich them through a waterfall, sell the audit trail.

  • Pluggable source adapters: Google Maps live, more queued
  • Waterfall enrichment with skip-already-filled to push match rates high
  • Per-value provenance: enricher, timestamp, confidence, cost
Visit →
E
Open source

Extremis

Drop-in memory for the Anthropic and OpenAI SDKs. Change one import; get persistent, learning memory.

  • Drop-in wrappers for `anthropic.Anthropic` and `openai.OpenAI`
  • Explainable recall — each result reports its score, age, and reuse count
  • Feedback loop re-scores memories so the useful ones surface first
Visit →GitHub
P
Open source

Peekr

Auto-instruments OpenAI and Anthropic clients and writes a JSONL span per call. Python + TypeScript.

  • One JSONL schema across the Python and TypeScript runtimes
  • Auto-instruments OpenAI and Anthropic SDKs — including streaming
  • Span tree propagated through async/await via AsyncLocalStorage
Visit →GitHub
F
In development

Friday

BDI agent runtime with four-layer memory. Biologically-inspired control modules over a vector store.

  • BDI-style (Belief–Desire–Intention) agent loop
  • Four-layer memory: episodic, semantic, procedural, identity
  • Biologically-inspired control modules over the memory plane
Visit →

Hosted Extremis and Peekr Cloud are in private testing — managed memory and observability for agent teams.

Join the waitlist →

Smaller tools

Compare CompaniesDrop in two or more offers. An agent researches each company — Glassdoor, Crunchbase, levels.fyi — scores them across 10 career parameters, and gives you a straight pick.
Writing

Engineering notes from the work.

What I learned building agent memory, running AI in production, and moving large search backbones — written up properly.

Aug 15, 2026
How I think about agent memoryMost LLM agents are amnesiacs. The fix isn't a bigger context window — it's a memory system with four layers, explainable retrieval, and a feedback loop.6 min readAISystemsExtremis
→Aug 8, 2026
How I think about agent observabilityTraditional APM was built for web requests, not for agents that loop, retry, branch, and spend dollars per call. Here's what agent observability actually needs — and why I built Peekr.6 min readAISystemsPeekr
→Aug 1, 2026
The AI agent observability stack — what to measure, what to useMost teams reach for one observability tool and call it solved. Different tools measure different things — traces, hallucinations, cost, drift. Here's a map of the space and where to start.6 min readAIObservabilityTooling
→Jul 25, 2026
Building a Brain: How I Gave My AI Assistant a Cognitive ArchitectureMost AI assistants are stateless. I wanted one that remembers, learns, and adapts — so I built Friday a brain.3 min readAICognitive ArchitectureReinforcement Learning
→Jul 15, 2026
What Every Backend Engineer Should Know About AttentionRNNs forced you to wait for token 100 before processing token 101. Transformers parallelize the whole sequence. Here's why that matters for production systems.5 min readAITransformersArchitecture
→Jul 4, 2026
RAG in Production: The Failure Modes Nobody Puts in the Pitch DeckRAG is in every AI pitch deck. The failure modes rarely are — retrieval collapse, frozen encoders, approximate MIPS. Here's what actually breaks it in production.7 min readAISystemsProduction
→
All 39 posts →

That archive also holds 30 paper breakdowns — a running series on the systems and ML papers behind the tools everyone uses.

Get in touch

Hard problem, real system, small team — let's talk.

Open to senior IC & tech-lead roles in distributed systems and applied AI. Remote or Dubai-hybrid, open to relocation. Also happy to talk shop with anyone building agent products.

Emailashwanijha04@gmail.comCalendar30-minute intro callLinkedIn/in/ashwanijha04GitHub@ashwanijha04CVDownload PDFX@ashwajha