Open to Software Engineering roles · Boston, MA

I build systems
end to end.

Software engineer with 4+ years across full-stack, AI systems, distributed systems, and cloud infrastructure — from a Kafka broker written from scratch in Go to a multi-AZ EKS platform running a custom Kubernetes operator and a self-hosted LLM.

Full-Stack AI Systems Distributed Systems Cloud & Infra
View Resume See Projects
Sai Vivekanand Reddy Vangala
01 — Focus

What I Work On

Four domains, each with something shipped behind it.

01 Full-Stack

  • React + TypeScript analytics SPA turning raw provider usage metadata into project-level dashboards
  • One React Native / Expo codebase shipped to both iOS and Android with signed EAS releases
  • Spring MVC + JSP enterprise frontend — cut forecast workbench load times 40%

02 AI Systems

  • Provider-agnostic LLM abstraction swapping Claude for local Ollama behind a single env variable
  • Distilled DeepSeek-V3 into a Qwen 0.6B student — Text-to-SQL accuracy 36% → 74%
  • Gemini Vision OCR pipeline at ~95% extraction accuracy across 10+ document formats

03 Distributed Systems

  • Kafka-compatible broker written from scratch in Go — binary wire protocol, correlation IDs, KRaft metadata
  • Kafka pipeline streaming 240k+ CVE records into PostgreSQL for downstream retrieval
  • PyTorch DDP over NCCL on 4 GPUs — 3.7× faster CNN training

04 Cloud & Infrastructure

  • Custom Kubernetes operator in Go (Kubebuilder) reconciling releases via CRDs and finalizers
  • Multi-AZ EKS provisioned end-to-end in Terraform — VPC, IAM/OIDC, KMS, Istio, Prometheus/Grafana
  • Containerized rollouts on GCP holding ~99.9% availability under monitoring
02 — About

About Me

I'm a software engineer with 4+ years of hands-on experience, currently building LLM infrastructure at IpserLab. I completed my MS in Computer Software Engineering at Northeastern University in December 2025.

What I actually enjoy is work that spans layers. I've hand-parsed a Kafka broker's binary wire protocol in Go, provisioned the multi-AZ EKS platform such a service would run on, trained and distilled the models sitting on top of it, and shipped the React front end a user finally touches. Enterprise scale at Blue Yonder, zero-to-one as a founding engineer at Puddl, and research-grade AI infrastructure at Humanitarians AI.

Education

Northeastern University
M.S., Computer Software Engineering
Boston, USA
Sept 2023 – Dec 2025
BITS Pilani
B.E., Electronics Engineering
Pilani, India
Aug 2016 – June 2020

Technical Skills

Languages & Frontend

PythonGoJavaC++TypeScriptJavaScriptSQLReactReact Native

Backend & Distributed Systems

KafkaSpring BootNode.jsFastAPIRedisPostgreSQLOpenSearchTCPConcurrency

AI / ML Systems

PyTorchDDPAMPCUDANCCLLangGraphRAGOllama

Cloud & Infrastructure

KubernetesDockerTerraformAWS (EKS)GCPHelmIstioPrometheusGrafanaCI/CD
03 — Experience

Work Experience

Software Engineer Intern
IpserLab ↗ Remote, USA
April 2026 – Present
  • Built a provider-agnostic LLM abstraction layer that swaps Claude for local Ollama models behind a single environment variable, so AI parsing isn't locked to one vendor
  • Prevented abuse of auth, OCR, and AI-suggestion routes with a Redis-backed sliding-window rate limiter that falls back to in-process limiting if Redis becomes unavailable, keeping the API online
  • Shipped one React Native / Expo codebase to iOS and Android, closing platform gaps like date pickers and camera permissions, and signed releases through EAS build profiles
AI Engineer Co-op (Backend & Infrastructure)
Humanitarians AI ↗ Boston, USA
Jan 2025 – May 2025
  • Owned ingestion, chunking, and metadata-aware ChromaDB indexing for a Gemini Vision OCR pipeline reaching ~95% extraction accuracy across 10+ document formats
  • Designed the state-passing contracts and episodic-memory indexing that multi-step agent workflows relied on, with OpenSearch-backed audit logs for traceability
  • Automated containerized rollouts on GCP with GitHub Actions, Terraform, and Kubernetes; health-check-based validation held availability at ~99.9%
Founding Engineer (Volunteer)
Puddl Bangalore, India
June 2022 – Aug 2023
  • Built a React / TypeScript analytics SPA that turned raw provider usage metadata (cost, latency, tokens) into project-level dashboards, syncing sanitized summaries to a Node.js backend for persistence
  • Kept provider API keys in-browser with a privacy-first client-side credential flow, so the backend never received or stored raw credentials
Software Engineer (Full-Stack)
Blue Yonder ↗ Hyderabad, India
July 2020 – May 2022
  • Reduced forecast workbench load times 40% for enterprise retailers by replacing greedy data fetching with a dynamic SQL query builder in the Java Spring backend
  • Added a server-side session-caching layer so planners could run rapid "what-if" simulations without repeated database writes, lifting forecasting throughput 25%
  • Fixed 70+ customer-reported bugs a year across the Spring MVC backend, JSP/JS frontend, and Oracle 18c queries, backed by JUnit regression suites holding 95%+ coverage
04 — Projects

Projects

Selected work. Every link goes to source you can read.

Cloud · Distributed Systems · AI 2026

Kubernetes Platform on EKS Featured

  • Custom Kubernetes operator in Go (Kubebuilder) that reconciles hourly CVE releases into processing Jobs using CRDs and finalizers
  • Kafka pipeline streaming 240k+ CVE records into PostgreSQL, feeding a RAG service backed by a self-hosted Llama3:8B
  • Multi-AZ EKS provisioned end-to-end in Terraform (VPC, IAM/OIDC, KMS) with Istio service mesh, Helm releases, Jenkins CI/CD, and Prometheus/Grafana plus Fluent Bit to CloudWatch
Go Kubernetes Terraform Kafka EKS Istio
View on GitHub →
Distributed Systems 2026

Kafka From Scratch

  • Kafka-compatible broker built from scratch in Go — length-prefixed binary request parsing, correlation IDs, ApiVersions and Fetch APIs
  • Concurrent client handling over a goroutine-based TCP server
  • Metadata-discovery and read/write protocol flows parse KRaft cluster metadata and pass the CodeCrafters test harness
Go TCP KRaft Concurrency
View on GitHub →
AI / ML 2026

Text-to-SQL Distillation Engine

  • Distilled DeepSeek-V3 outputs into a compact Qwen 0.6B student model
  • Improved schema-grounded Text-to-SQL execution accuracy from 36% to 74% on a held-out query set
  • Cut serving latency and cost by deploying the quantized student through Ollama GGUF, enabling local SQL generation over a Pandas/SQLite schema layer with no remote teacher inference
Python Qwen Ollama DeepSeek-V3
View on GitHub →
HPC / ML 2025

Distributed Training Systems

  • Accelerated CNN training 3.7× at 92% scaling efficiency — 19,140 samples/second — with PyTorch DDP over NCCL across 4 GPUs on a single node
  • Cut GPU training time 66% and memory footprint 31% while retaining 93% accuracy by integrating Automatic Mixed Precision with bfloat16 autocast
  • Parallelized CPU preprocessing with Joblib and Dask ahead of the GPU pipeline
PyTorch DDP NCCL CUDA Dask
View on GitHub →
AI / Agents 2025

Multi-Agent Research Assistant

  • LangGraph agents orchestrate web and academic retrieval, then synthesize cited findings into a single answer
  • Airflow and Docling ingestion pipeline into Pinecone, with a FastAPI backend and web UI
  • Two-person team project — design and implementation split equally with a teammate; repo forked from their original
LangGraph Airflow Pinecone FastAPI
View on GitHub →
AI / Data 2024

Multimodal RAG Ingestion Platform

  • Automated recurring ingestion of CFA publications by orchestrating Selenium browser workflows with Airflow, extracting PDFs, images, titles, summaries, and metadata into AWS S3 and Snowflake
  • Built backend retrieval workflows with FastAPI, document chunking, vector indexing, and OpenAI/NVIDIA services to support multimodal Q&A, summary generation, and report creation
Selenium Airflow Snowflake FastAPI AWS
View on GitHub →
Cloud / Backend 2024

Auto-scaling User Service on GCP

  • Decoupled account creation from email delivery with an event-driven Pub/Sub flow, handing verification off from the Spring Boot API to an async Cloud Functions consumer
  • Provisioned a self-healing, autoscaling GCP environment with Terraform: CMEK-encrypted Cloud SQL/VM/storage and a Packer-built instance group (3–6 replicas) behind an HTTPS load balancer with health-check auto-recovery
Java Spring Boot Terraform GCP Packer
View on GitHub →