SREENJAY SEN
Backend Engineer
Pune
Summary
Backend Engineer with ~3 years of experience designing, building, and operating production distributed systems in Java 21 and Spring Boot. Strong foundation in hexagonal architecture, event-driven design (CQRS, Domain Events, AWS SQS/SNS), and high-reliability cloud services on AWS. Differentiated by hands-on, production-grade ML infrastructure experience — RAG pipelines, hybrid vector retrieval, cross-encoder reranking, on-GPU LLM serving via vLLM, and parameter-efficient fine-tuning (LoRA/QLoRA). Comfortable owning systems end-to-end and mentoring engineers across the stack. MSc Data Science (HSE Moscow); published ML researcher.
Overview
3
3
years of professional experience
Work History
Backend Engineer
UpThink Edutech Pvt. Ltd.
04.2023 - Current
- Designed core backend services as a hexagonal-architecture Spring Boot system with package-per-bounded-context separation, cleanly extractable into microservices as the platform scales.
- Operated Spring Boot + PostgreSQL services on AWS (Lambda, RDS, S3, Cognito); used Cloud Watch for real-time monitoring, alerting, and incident triage.
- Containerized services with Docker; managed configuration and secrets via AWS Parameter Store with Spring profile-based environment isolation.
- Built GitHub Actions CI/CD pipelines: automated builds, JUnit 5 + Testcontainers integration tests, and gated deployments across dev, staging, and production.
- Event-Driven Architecture & Distributed Messaging
- Designed event-driven workflows on AWS SQS for decoupled async processing of long-running jobs (notifications, scheduled tasks, ML inference handoffs) and AWS SNS for fan-out to downstream consumers.
- Implemented CQRS with DDD — separate command and query models, domain events via Spring ApplicationEventPublisher, and SQS-backed async handlers for write-side operations.
- Built reliability into every async flow: dead-letter queues (DLQ), idempotency guards, exponential backoff, and structured error handling.
- Centralised Identity & Access Management Platform
- Architected and built a standalone IAM service as the authentication and authorisation backbone for the entire product suite, using JWT token introspection, RBAC, and Spring Security.
- Implemented Google SSO anchored on the stable sub claim, with an employee_email_history table preserving identity continuity across email changes.
- Delivered dual auth (Google OAuth 2.0 + password), account linking, identity deduplication, and concurrent session management.
- Production ML Infrastructure — RAG Pipeline & LLM Serving
- Designed and shipped an end-to-end RAG pipeline powering a production AI tutoring chatbot used by 200+ Subject Matter Experts as a tutoring assistant across mathematics, statistics, chemistry, biology, physics, business, and computer science.
- Engineered hybrid retrieval combining BGE-M3 dense and sparse embeddings against a Qdrant vector store with Elasticsearch fallback, sustaining sub-200 ms P95 retrieval latency under production load.
- Integrated a cross-encoder reranking stage (mxbai-rerank-v2-base) on top of first-stage retrieval, measurably improving answer relevance on internal evaluation sets.
- Deployed Qwen-3.5B via vLLM on GPU for high-throughput inference; consolidated HyDE into the primary prompt path, reducing pipeline complexity without quality regression.
- Applied parameter-efficient fine-tuning (LoRA / QLoRA) on the inference model for domain adaptation to tutoring content; improvements validated through Subject Matter Expert review across mathematics, sciences, and humanities.
- AI Essay Grader — Few-Shot RAG Pipeline for Rubric-Based Evaluation
- Designed and shipped a two-stage essay grading system used in production by 50+ English Subject Matter Experts as a human-in-the-loop tool — the system produces a first-pass evaluation, SMEs review and refine — across 40 distinct essay categories.
- Stage 1: built an essay-type classifier that routes each submission to one of 40 rubric categories (narrative, argumentative, expository, analytical, etc.).
- Stage 2: implemented few-shot RAG — retrieved 2-3 human-graded exemplar essays from the matched category, then prompted the LLM to grade the submission against that category’s rubric using the exemplars as in-context anchors.
- Produced structured, multi-dimensional feedback covering organisation, development, narrative elements, style/voice, and grammar/mechanics, with inline annotations on the student’s text.
- Engineered the grading pipeline as an async SQS-backed workflow with DLQ and idempotency, isolating long-running LLM calls from the request path.
- Mentorship & Collaboration
- Mentored a team of 5 engineers; contributed to architecture and code reviews, and collaborated cross-functionally on the Upthink CMS.
Education
MSc - Data Science
08-2022
- Thesis: Predicting arterial blood pressure from non-invasive 4D flow MRI data using physics-informed Generative Adversarial Networks.
- Publication: “ML models for money demand forecasting in the Indian economy”, HSE Economic Journal, 2024.
- GPA: 7.8 / 10
B.Sc. - Electronics & Communication Engineering
05-2013
GPA: 7.15 / 10
Skills
Languages & Frameworks: Java 21, Spring Boot 3, Spring Security, Spring Framework, Hibernate ORM, JPA, Python, JavaScript, SQL
Architecture & Patterns: Hexagonal / Clean Architecture, Domain-Driven Design (DDD), CQRS, Event-Driven Architecture, Microservices, RESTful APIs, SOLID Principles, OOP, Concurrency
Distributed Systems & Cloud: AWS (Lambda, RDS, S3, SQS, SNS, Cognito, CloudWatch, Parameter Store), Docker, GitHub Actions CI/CD, Gradle, Git
Databases & Storage: PostgreSQL, Redis, Qdrant (vector DB), Elasticsearch, JDBC, Query Optimisation
Timeline
Backend Engineer - UpThink Edutech Pvt. Ltd.
04.2023 - Current
Heritage Institute of Technology - B.Sc., Electronics & Communication Engineering
National Research University Higher School of Economics (HSE) - MSc, Data Science