sumedha@portfolio:~$ whoami
Sumedha
Pandravada
Pandravada
Backend Engineer
AI / ML Dev
Cloud & Systems
@ Oracle · MTS-1
Building backend and AI/ML systems 🖥️
I enjoy coding in Python and have built most of my projects in it.
At Oracle I worked on MySQL Server core (C++) and MySQL Database Service on OCI.
I like turning AI ideas into production-grade software, especially systems involving
LLMs, RAG pipelines, and agentic workflows.
Overall, I work at the intersection of backend infrastructure and AI engineering.
01 / about
About Me
// who I am · what I do · where I build
Hi, I'm Sumedha 👋 I'm a software engineer interested in backend systems, distributed infrastructure, and AI engineering. At Oracle, I worked on MySQL Server core and MySQL Database Service (OCI), where most of my time went into debugging production issues, fixing server bugs, and improving reliability. Outside work, I build a lot of AI projects — usually involving LLMs, multi-agent systems, RAG pipelines, or developer tooling. I enjoy building products that are actually useful — either to real users or to me as a developer. I'm a fast learner and like picking up whatever is needed to get things working end-to-end.
Education
🎓 Mahindra University
2021 – 2025
Bachelor of Technology in Computer Science
CGPA: 9.05 / 10 · Academic Merit Scholarship — INR 4,00,000
🔬 National University of Singapore (NUS)
Jul 2023
Academic Research Programme
Built a YOLOv8 safety gear detection — 80% precision over 100 epochs (team of 5)
02 / experience
Experience
Oracle
Member of Technical Staff - 1
C++MySQLPythonOCIMCPLLMs
Fixed sev-2/3 C++ bugs in MySQL Server core, backporting patches across MySQL 8.0, 8.4, and trunk with full MTR validation.
Mitigated multiple severity-2 to severity-4 OCI MDS on-call incidents spanning OOM failures, networking faults, and high-availability outages.
Wrote a Python automation script processing 30–40 JIRA tickets/week for a 15-member team, saving manual triage effort per week.
Configured a MCP-connected LLM agent workflow in Codex to accelerate MySQL bug triage and root-cause analysis.
Axtria
GenAI Intern
LangChainFAISSAzure OpenAIPythonRAG
Built an internal RAG Document Q&A System — tested across 50+ documents with optimised chunk size, overlap, and top-k retrieval for answer relevance.
Developed LLM-powered marketing email generator (prompt pipelines over Excel, CSV, PDF) and a natural language data visualization feature generating Matplotlib code.
Fanztar
Full Stack Intern
PythonVue.jsGrapeJSboto3AWS S3MongoDB
Reduced content creation time by ~67% via GrapeJS visual editor integration.
Automated PDF generation and a boto3 pipeline to upload images/PDFs to Amazon S3 and persist URLs in MongoDB.
03 / skills
Skills
Languages
PythonC++SQLJavaJavaScript
Backend
FastAPIFlaskREST APIsSystem DesignDistributed Systems
AI / ML
LLMsRAGLangChainLangGraphMCPFAISSAgents
Cloud / DevOps
AWSOCIKubernetesDockerCI/CD
Databases
MySQLPostgreSQLMongoDBRedisSupabase
Frontend
ReactVue.js
04 / projects
Projects
// Project: description, stack, highlights, link
Budget Manager Bot
Multi-agent personal finance assistant (Telegram)
LangGraphFastAPIPostgreSQLMCPLLMs
github.com/sumedha0108/budget-agent
- LLM supervisor routing & chaining 5 specialized agents and 10+ financial tools
- 18+ tested workflows; exposed as MCP server — Claude Desktop native
Job Tracker & ATS Analyzer
Full-stack job tracker with LLM ATS resume scoring and Mock Interview
FastAPIReactPostgreSQLKubernetesGroq
github.com/sumedha0108/job-tracker
- llama-3.3-70b resume scoring — match score, missing keywords, AI-rewritten summary
- Voice mock interview system with 4-second silence detection & LLM evaluation
- K8s deploy: HPA (1–5 replicas at 70% CPU), Prometheus/Grafana monitoring across 9 pods
Distributed Task Queue
Multi-node async task queue
PythonRedisFastAPIDocker
github.com/sumedha0108/task-queue
- 4 async job types, real-time status tracking, exponential backoff retry logic
- 5-thread ThreadPoolExecutor worker pool achieving ~40% faster throughput
Research Agent
Multi-agent deep research system
PythonOpenAI Agents SDKOllamaGradio
github.com/sumedha0108/deep-research-agent
- 3-agent system (planner, search, writer) using OpenAI Agents SDK
- Planner runs 5 parallel DuckDuckGo searches, each summarised into 300-word digests
- Produces 1,000+ word Markdown reports via a Gradio interface
FinSight
Real-time platform for tracking anomalies across portfolio assets
PythonFastAPIReactSupabaseGoogle Gemini
- Live price data for 15+ stocks & crypto assets every 60 seconds
- Z-score anomaly detection across 3 severity tiers (low / medium / high)
- Gemini 2.5 Flash + Ollama failover for AI-generated anomaly explanations
- Sentiment pipeline with 6-hour caching across 5 news headlines per asset
06 / contact
Get in Touch
// Always open to interesting problems & collaborations
Whether it's a challenging engineering problem, an interesting AI project, or just a conversation about distributed systems and LLMs — reach out anytime.