Sriram Vivek

< />

Building reliable backend systems and AI-powered pipelines — from event-driven payment infra to retrieval-augmented agents.

About

I enjoy building backend systems that hold up under real conditions.

I've designed a microservices backend with gRPC, adding circuit breakers, mutual TLS, and full OpenTelemetry/Prometheus/Grafana observability to see exactly how the system behaves when things fail, not just when they work. I've also built an event-driven payment processor on Kafka and FastAPI, with idempotent event handling and a dead-letter queue so failures are recoverable instead of silent.

More recently, I've been getting hands on with AI infrastructure. During my internship at Galatea Associates, I built a retrieval pipeline for a major financial services client, turning a 600-page technical spec into context an LLM could reliably use for narrative generation. I've also built a full retrieval-augmented system from scratch, including a custom MCP server as the sole access layer to the underlying data, and deployed it end to end with containerization and CI/CD.

I'm looking to apply these skills somewhere I can dig into hard systems problems and make a real impact.

Feel free to reach out or connect, always happy to talk shop.

Experience

Financial Software Engineer Intern · Galatea Associates LLC

June 2025 – August 2025

Boston, MA

Built a RAG service as part of a development-only AI proof of concept for a financial services client, ingesting a 600-page portfolio optimizer specification to supply retrievable context for LLM-generated trade narratives.
Built the document ingestion pipeline: PyMuPDF for text and table extraction, LangChain's recursive character splitter for chunking, and text-embedding-3-large for embeddings, stored in a PGVector store for similarity retrieval.
Wrote SQL against MS SQL Server to extract and reshape portfolio positions, constraints, and trade outputs into structured tables for LLM prompt consumption.
Authored technical documentation in Confluence and tracked deliverables in JIRA under Agile methodology.

Skills

Drag them around — these are the top skills I work with as an Engineer!

Python

FastAPI

gRPC

PostgreSQL

Apache Kafka

Docker

Nginx

GitHub Actions

Prometheus

Grafana

Jaeger

MCP

Qdrant

Next.js

TypeScript

My Work

Agentic RAG System

An end-to-end RAG pipeline over Anthropic's API documentation — markdown-aware chunking into ~3,100 chunks, sentence-transformer embeddings, and a self-hosted Qdrant vector store, with full ingestion completing in under 60 seconds. Retrieval is exposed through a custom MCP server (built on the Python MCP SDK) with 4 tools, wired into an agent framework so the agent can only retrieve through MCP — no direct database access. The whole stack runs on a self-managed VPS behind Nginx/TLS with Docker, systemd, and a scripted GitHub Actions redeploy, with a Next.js frontend that includes a retrieval-inspector UI showing citation sources and similarity scores.

Why it's interesting: The interesting part is the enforcement boundary: the agent is architecturally incapable of bypassing the MCP layer to hit the vector store directly, and every answer traces back to its source chunks with a similarity score.

Python
MCP
Qdrant
FastAPI
Docker
Nginx
Next.js
TypeScript
GitHub Actions

Read full write-up View on GitHub

Real-Time Event-Driven Payment Processing Backend

An event-driven payment backend using FastAPI and Apache Kafka to decouple synchronous API ingestion from asynchronous downstream processing, backed by a PostgreSQL state machine that tracks each payment through pending, processed, and failed states. Processing is idempotent — event IDs are tracked with conflict-safe inserts and Kafka offsets are committed manually, so a consumer crash can't cause duplicate charges or lost events. Failed events are routed to a dedicated dead-letter queue topic with structured error context, validated by load-testing with a producer simulating 1,000 payment requests.

Why it's interesting: The focus here is correctness under failure — idempotency, manual offset commits, and a DLQ are the unglamorous parts of payment infrastructure that actually matter once something crashes mid-stream.

FastAPI
Apache Kafka
PostgreSQL
Python

Read full write-up View on GitHub

Distributed Microservices Orchestration & Resilience Engine

A multi-service gRPC backend with a central orchestrator aggregating responses from separate User and Search services over protobuf-defined contracts, using server-side streaming for real-time result delivery. Circuit breakers and exponential backoff retries contain partial failures across services, and mutual TLS secures every cross-service call. Distributed tracing runs through OpenTelemetry gRPC interceptors capturing both function- and business-level spans, visualized in Jaeger, with Prometheus and Grafana for metrics — the observability stack runs via Docker Compose.

Why it's interesting: Built to understand, hands-on, how the resilience and observability patterns behind real service meshes actually behave — what a circuit breaker looks like mid-trip, and what a trace looks like when a downstream service is failing.

gRPC
Protobuf
OpenTelemetry
Jaeger
Prometheus
Grafana
Docker Compose

Read full write-up View on GitHub

Contact

sriramv1202@gmail.com