Sirisolla Jaya Surya
Backend & AI Agent Engineer
4 years building production distributed systems at scale (Reliance Jio, IPL 2025 live broadcasts) and AI/agent platforms (OpenClaw + MCP, on-device LLMs). Go + Kafka + Flink for the data plane; LLM tool-calling, RAG, and verifiable-reward training (GRPO) for the agent layer.
# Tech Stack
A focused stack — the tools I reach for daily to ship production systems and AI agents.
How I build APIs, services, and the dashboards on top of them.
Where the data lives, how it moves, where it runs.
The agent layer — tool-calling, retrieval, on-device inference, verifiable-reward training.
# Experience
4 years shipping distributed backends and AI agents in production.
Senior Engineer — Backend Development
Quest Global · Client: Reliance Jio
Architect of a fault-tolerant backend ecosystem and observability stack for Reliance Jio — powering high-traffic workflows including IPL 2025 live broadcasts. Building the agent layer on top with OpenClaw + MCP.
Distributed Systems Architecture
Fault-tolerant backend + Next.js dashboards in Go / gRPC, sustaining 99.9% availability across high-traffic workflows.
OpenClaw + MCP Agent Platform
Exposed project services and Elasticsearch log analytics as MCP tools. WhatsApp & RCS channels let stakeholders query status and pull logs from any device.
Event-Driven Pipelines
Real-time streaming with Apache Flink + Kafka, feeding processed telemetry into React dashboards with sub-second visibility.
IPL 2025 Critical Event Management
Custom log analytics engine + observability views — sharpened Root Cause Analysis during the IPL 2025 live streaming season.
LLM-Powered Incident Diagnostics
Integrated Mistral 7B into the observability pipeline to auto-summarize error clusters and suggest probable causes from production logs.
Security Engineering
Role-Based Access Control (RBAC) and encryption middleware enforced across both backend APIs and frontend access layers.
Associate Software Developer
Udify Technologies
End-to-end backend ownership across two production platforms — greenfield Healthcare Management and a FinTech microservice refactor.
Healthcare Management System — Greenfield
- Designed DB schema and full backend + frontend workflow from scratch with Node.js and PostgreSQL.
- Used Sequelize ORM to ensure strict data integrity and efficient query performance.
- Implemented RBAC to secure sensitive patient data in line with healthcare compliance standards.
- Architected a secure Razorpay + wallet payment service with zero transaction failures for high-volume operations.
FinTech Microservice — NestJS + MongoDB
- Owned an existing FinTech service; refactored NoSQL data models and backend logic.
- Improved scalability and data retrieval speeds by 30%.
- Hardened service boundaries and observability for production traffic.
# Projects
Things I've shipped outside of work — mostly agent systems, on-device LLMs, and side products.
LogSentinel v2
Multi-Agent SOC Simulator · RLVR / GRPO
Multi-agent SOC simulator for training LLMs on long-horizon incident response with verifiable rewards. 4-role agent team over a 5-phase DETECT→TRIAGE→MITIGATE→VERIFY→FINAL_REPORT lifecycle, with a latent world-state model that only changes under evidence-backed mitigations.
- Trained Qwen2.5-7B via GRPO (TRL + Unsloth) with adaptive curriculum.
- Shipped as an OpenEnv-compliant FastAPI environment for reproducible runs.
- Composite reward (outcome + detection F1 + severity + efficiency + teamwork) with anti-reward-hacking penalties.
- Benchmarked 0.49 avg reward vs 0.41 heuristic baseline; resolves incidents in 8 steps vs 45.
On-Device Mobile Agent
Koog + llama.cpp · Fully Offline
Fully offline personal mobile agent built on the Koog framework with llama.cpp for on-device inference. Queries the device messaging database via tool calls to summarize conversations, surface missed deals or important messages, and track personal spending — no cloud round-trip.
- Benchmarked SmolLM2 (350M/700M), LFM2.5 (350M/700M), and Qwen2 0.5B for tool-calling latency and accuracy on mobile hardware.
- Tool-call layer over local SMS / conversation database for summarization and search.
- Privacy-first architecture — zero data leaves the device.
MaMenu
QR-Based Restaurant Ordering
Side product — a contactless digital menu and hotel operations platform. Go backend with a Next.js frontend, full RBAC, and end-to-end table and order management.
- Go backend + Next.js frontend, deployed as a single ops platform.
- End-to-end table, order, and menu management flows.
- Role-Based Access Control across staff, kitchen, and admin roles.
# Certifications
Continuing-education credentials in agentic AI, prompt engineering, and core CS — alongside Scaler skill-mastery assessments.
Agentic AI for Beginners
Udemy · Aakriti E-Learning Academy
Prompt Engineering for Everyone — Bootcamp
Udemy · Navid Shirzadi, PhD
# Education
Scaler Academy
Software Development — System Design, Distributed Systems, DSA
Curated pattern-based notes for all LeetCode Top 150 interview questions.
leetcode-top-150SRKR Engineering College
Bachelor of Engineering (B.Tech)
# Achievements
Partner Award
Reliance Jio
Recognized for technical excellence and delivering high-impact solutions.
Rising Star Award
QuEST Global
Awarded for rapid career progression and consistent high performance.
On-the-Fly Spot Awards (×2)
QuEST Global
Two spot awards for critical contributions to production releases.
# Let's Connect
Open to backend, distributed systems, and AI/agent engineering roles. Got a problem in observability, real-time streaming, or LLM tool-calling? Let's talk.
suryasirisolla@gmail.com
Phone
+91 94906 47918
jaya-surya-sirisolla
GitHub
surya-sirisolla
Location
Hyderabad, India