🥑
[rag]knowledge base········OK
[vec]bge-base · ONNX········OK
[llm]gemini-2.5-flash········LIVE
[sys]avocado online········✦ ready
>
Open to work · Full-time · Remote · Hybrid
Hey, I'm Jaya. I build things, then I keep them running.
The first thing I ever shipped was a one-page site for my dad's brick business, in a mid-sized town in India. It brought him real customers during COVID. Nothing I've built since has felt more like the point.
These days that's AI infrastructure: retrieval pipelines, inference at the edge, telemetry that can't drop a byte. Same job, bigger machines.
Kavali, Nellore2020
Why I build
Where this started, and why it still decides what I pick up.
1 named user · HTML + SEO · live since Jun 2020
In 2020 my father started a brick business in our town. I built him a one-page website. HTML, some JavaScript, and a stubborn amount of SEO. No framework, no analytics, no plan.
People started finding him through it. During COVID, that little page was doing real work.
I've since built systems that carry 115GB a day and answer 3,000 requests a second. None of them taught me more than that one did: software is at its best when it's the shortest path between someone's need and someone's help. Everything on this page is downstream of that.
"Do hard things" isn't about difficulty for its own sake. It's about not abandoning something the moment it stops being fun.
Vellore · VIT2020
gradeVITian
Built as an undergrad, still on call. Six years live and five graduating classes later.
~17,000 people a month · sub-second mobile loads · on call since Aug 2020
I built gradeVITian while I was still an undergrad, because my friends and I kept doing the same grade math on paper the night before results came out.
I graduated. They graduated. It's still up. Five classes have used it now, 2020 through 2025. That's the part I'm actually proud of. The traffic followed on its own.
The grade & attendance calculators tens of thousands of VIT students reach for every exam season. Designed, shipped, and still run by me.
A high-traffic progressive web app for academic grade forecasting: GPA, CGPA, grade prediction, and attendance, with sub-second mobile loads and #2 Google ranking from programmatic SEO. Six years live, rebuilt on Next.js + FastAPI.
Milestones & achievements in the gallery · free, no sign-up at gradevitian.jayaremala.com
17K+
Monthly active users
20K+
Registered accounts
#2
On Google Search
6+ yrs
Live in production
Postscript, 2026
After six years in production and 17,000+ monthly users, gradeVITian has been rebuilt from the ground up, re-engineered on Next.js and FastAPI and relaunched at gradevitian.jayaremala.com. Heartfelt thanks to the VITian community (Class of 2020–2025) for making the original such a success; this journey defined me as an engineer, and this new chapter carries it forward.
Offshore · Shell PLC · NYU2021—present
The work
Production systems, and the numbers they hold up under.
200+ offshore stations · 115GB/day at zero data loss · 3K+ RPS at 99.9% uptime
When it has to hold.
0%
Latency Cut
P99 RAG on 3K+ RPS
0K+
Peak RPS
99.9% uptime
0ms
Edge Inference
Snapdragon NPU
0GB
Daily Throughput
Zero data loss
0%
Latency Cut
P99 RAG on 3K+ RPS
0K+
Peak RPS
99.9% uptime
0ms
Edge Inference
Snapdragon NPU
0GB
Daily Throughput
Zero data loss
I build AI systems that work under real production pressure. At NYU, I cut RAG query latency by 78% on a Multi-Agent research engine serving 3K+ RPS, and pushed LLM inference to 15ms on Snapdragon NPUs via QLoRA + AWQ quantization. Before that, I kept Shell's maritime telemetry alive 115GB/day, 200+ offshore stations, zero data loss.
The resume version
Software Engineer with 5+ years of experience building production-grade AI infrastructure at the intersection of Distributed Systems and Agentic AI. Designed high-throughput microservices sustaining 3K+ RPS at 99.9% uptime, cut P99 RAG latency by 78% on enterprise research pipelines, and pushed LLM inference to 15ms on Edge NPUs. Proven across Fortune 500 energy infrastructure (Shell PLC) and research-scale AI systems (NYU IT). Qualcomm Edge AI Hackathon Winner.
Featured Projects
All projects →Tailorbird
An AI agent that tailors resumes for every role, on autopilot which stitches leaves into a nest, this one stitches your experience to fit each job. An end-to-end system that turns a job-posting URL into a tailored, one-page resume (PDF + DOCX) with factual integrity, and does it autonomously at scale. You paste URLs into a local web tracker, click Run agent, and come back to finished resumes, dropped roles (with reasons), and a short list of judgment calls for you to decide. What inspired it? 1. The tailoring method comes from the open-source /resume-tailoring Claude Code skill (MIT, by Varun R). Tailorbird wraps that skill in automation rather than reinventing it, so quality equals a careful, hand-guided session. 2. The agent-over-skill design (deterministic Python for the mechanical 80%, the model only for authoring) came from watching where tokens and time actually went, and from the skill's own "token optimization" guidance. 3. The human-in-the-loop needs_review valve exists because some calls (an unusual level, a missing must-have, a borderline fit) genuinely need a person, so the agent parks those instead of guessing. Who benefits. 1. High-volume job seekers - apply to many roles without the per-role grind; a batch that took an evening now runs in minutes while you do something else. 2. Career switchers - the archetype/reframing engine surfaces transferable experience for roles adjacent to your background. 3. Non-native resume writers - get clean, idiomatic, role-matched phrasing without fighting the blank page. 4. Busy people - schedule it, results and a review list are waiting when you're back. 5. Anyone who values honesty - truthfulness boundaries are enforced; it will flag a poor fit rather than pad it.
ClaudeCodeCLIPythonFastAPIAgents
jayaremala.com (Avocado AI)
Production portfolio with Avocado AI, a streaming RAG agentic chatbot backed by a 4-stage hybrid retrieval pipeline: query expansion (up to 4 variants), batched dense search via ChromaDB (all-MiniLM-L6-v2 ONNX), BM25 lexical search (rank_bm25), and Reciprocal Rank Fusion (k=60), all before Gemini 2.5 Flash sees the question. Implement model fallback mechanism via Gemini, Grow and OpenRouter with the knowledge base of ~80 atomic documents auto-syncs incrementally: new or edited documents are upserted, deleted documents are purged from ChromaDB, unchanged ones are skipped entirely. Runs as a FastAPI Docker container on AWS Lightsail with zero-downtime blue-green deployments, Nginx + Let's Encrypt HTTPS, daily S3 backup of SQLite analytics, and a Next.js 16 static frontend on GitHub Pages.
Next.js 16FastAPIPythonChromaDB
Qualcomm Hackathon Winner
SnapLog – Edge AI Security Engine
Achieved 15ms token latency on Snapdragon NPUs, a 10× improvement over cloud inference, by fine-tuning Llama 3.2 3B on security logs with QLoRA and deploying via 4-bit AWQ quantization through ONNX Runtime on-device. Guaranteed zero data loss during network partitions via an offline-first SQLite buffer with background sync workers.
PythonFastAPIQLoRAAWQ Quantization
Multi-Agent Research Discovery Engine
Production LangGraph + Llama 3.1 70B system that semantically maps global researcher collaboration networks by indexing millions of papers from Elsevier's Science Direct/Scopus. Cut P99 RAG latency by 78% with Write-Through Redis caching; sustained 99.9% uptime at 3,000+ RPS on AWS ECS.
LangGraphLlama 3.1 70BBGE-M3FastAPI
CodeCollab – Real-Time Collaborative Code Editor
Conflict-free simultaneous multi-user editing using Yjs (CRDTs) and WebSockets, scaled horizontally via Nginx load balancing across containerized instances. Increased AI auto-complete context quality by 65% with a Context-Aware Coding Agent using AST-based chunking and Voyage-Code-2 embeddings.
AWSClaude 3.5Voyage AIRedis
gradeVITian
High-traffic progressive web app for academic grade forecasting at VIT. Scaled to 17K+ monthly active users and around 20K+ registered accounts. Achieved #2 Google Search ranking via programmatic SEO with sub-second mobile load times.
Next.jsReactFastAPISEO
After six years in production and 17,000+ monthly users, gradeVITian has been rebuilt from the ground up, re-engineered on Next.js and FastAPI and relaunched at gradevitian.jayaremala.com. Heartfelt thanks to the VITian community (Class of 2020–2025) for making the original such a success; this journey defined me as an engineer, and this new chapter carries it forward.
VRF Bricks
The website for my father's fly ash brick yard in Kavali, Andhra Pradesh. First built during COVID to help people find the business, and rebuilt in 2026: bilingual in English and Telugu, WhatsApp-first enquiries, and local SEO so a customer twenty kilometres away can find the yard at all.
Next.jsReactTypeScriptTailwind
At its heart, this was a project of purpose: an online gateway built to help people navigate my father's new business within our town. Seeing a simple web application drive real-world sales and provide stability for my family during the COVID-era was a transformative experience. It taught me that engineering is most powerful when it serves a personal and communal need. Six years on, it has been rebuilt from the ground up and relaunched at vrfbricks.jayaremala.com. Bilingual in English and Telugu, because Kavali is a Telugu market. WhatsApp instead of a contact form, because that is how the construction trade actually talks. And the one real photograph of the yard is finally the first thing you see.
New York2026
What I optimize for
The reason the rest of this page exists.
hybrid retrieval · a public MCP server · live traces at /system
Still running
Things I shipped that are still in production. The counters are live.
All three are still up. The oldest has been on call since 2020.
Where I stand
Opinions held loudly enough to put on the page. Every line here is something already shipped, not a slogan.
What I'm for
Things that stay up
Two of these shipped in 2020 and are still in production. Neither one is a portfolio piece.
Meeting people in their own language
A Telugu market gets a Telugu site. That is the baseline, not a feature.
Deterministic where it can be
Plain Python handles the mechanical eighty percent. The model is only asked to author.
Being on call for what you shipped
Six years on, the calculators still answer during every exam season.
One real user before any of the rest
The first site had exactly one person in mind, and it worked because of that.
What I won't do
Padding a poor fit
Roles that don't match get dropped with a reason, instead of stretched until they do.
Guessing when a person should decide
Some calls genuinely need a human, so the agent parks them instead of guessing.
Reinventing what already works
Tailorbird wraps an existing skill in automation rather than rebuilding it from scratch.
A contact form where the trade uses WhatsApp
WhatsApp instead, because that is how the construction trade actually talks.
Stock photography standing in for the real thing
If the photograph is not of the actual yard, it does not go on the page.
How I work
Currently pointed at recommendation systems and search at scale & how they shape the way people discover information.
Currently, I'm obsessed with recommendation systems and the search technology at scale, where it powers the way humans think and behave with intention and responsibility.
Skills & Tools
Previously
NYU IT High-Speed Research Network, Shell, Wipro
Domains
EnergyBioinformaticsScientific Research
Excited in
Health CareFinanceConsumer Space (Entertainment and Retail)& more
Career path
NYU CAS
NYU IT
NYU
Wipro Limited
Vellore Institute of Technology
Skills in Action
scroll → Project Experience Award ★· hover a skill to see where it's been used
Kind words
What the people I actually worked with said, unprompted.
1 / 5
5 recommendations
Sabarish is an intuitive and intelligent person, who has offered me solutions in the most difficult times and helped me at multiple occasions. He is prompt, hardworking with a knack for technology, and an absolute delight to work with!
DC
LinkedIn
Dec 2022
Connect
Open to full-time roles, remote or hybrid. This is the fastest way to reach me.
Open to software engineering roles in AI/ML infrastructure, distributed systems, and full-stack. Healthcare, finance, and consumer products are where I'm most curious right now, but any hard problem will do.
Book a call
Thirty minutes, on Google Calendar. The times below are real openings from my week.
Checking the calendar…
Send a message
Not ready for a call, or have a question first? This goes straight to my inbox.