Ansh Saxena — Backend & ML Infrastructure Engineer at Cyware Labs, Bengaluru
I'm Ansh Saxena, a Backend & ML Infrastructure Engineer at Cyware Labs in Bengaluru, India. I build production distributed systems and LLM/agentic platforms in Python, Go and Java: multi-tenant services, event-driven pipelines, and RAG and multi-agent systems that run in production.
Backend & ML Infra Engineer @ Cyware Labs | Java · Go · Python | Kafka · Kubernetes · LangGraph/RAG | Bengaluru
About
I work where backend engineering meets applied AI. At Cyware Labs (B2B cybersecurity SaaS) I build backend services and LLM infrastructure in Python, Go and Java. In my own time I built an open-source SOC alert-triage agent on LangGraph, FastAPI and Pydantic, and a set of JMH benchmarks for low-latency Java (lock contention, a lock-free SPSC queue, allocation-free hot paths). I hold a B.Tech in Artificial Intelligence & Machine Learning from Vellore Institute of Technology (VIT).
Experience
Software Engineer, ML Infrastructure & Backend — Cyware Labs
Backend and ML infrastructure work in Python, Go and Java: multi-tenant services, event-driven pipelines, and LLM orchestration and agentic systems running on PostgreSQL, NATS JetStream and Kubernetes. Details of internal products are not public.
Software Engineering Intern — PreProd Corp
- Drove a 25% business-metric gain at 90% prediction accuracy by deploying a Gradient Boosting pricing model behind production REST APIs; sustained 99.9% uptime on a PostgreSQL ETL pipeline processing 2 GB+/day via SQL query optimization and pipeline redesign.
Machine Learning Engineer Intern — Omdena
- Cut cloud inference cost 30% and raised model accuracy to 92% by applying TensorFlow quantization to the served model on Docker + AWS at sub-350ms latency, with a 40% Airflow ETL throughput gain on a data-processing pipeline handling 100K+ records/day.
Projects
SOC Triage Agent (open source)
An agentic alert-triage service. It ingests alerts from Splunk, CrowdStrike, AWS GuardDuty and Sentinel (plus a generic format, with source auto-detection) and normalises them to OCSF. It then runs a LangGraph DAG with MITRE ATT&CK-routed enrichment and emits a structured TriageVerdict (Pydantic) via async HTTP 202 + polling. It also has a pluggable LLM backend, API-key auth, a Pytest suite and OpenAPI/Swagger docs.
An optional event-driven runtime publishes alerts to NATS JetStream through a write-ahead log with a reconciler, so an alert survives a broker outage. Service-to-service auth uses JWTs, and analysts get live updates over WebSockets held at the edge by a Cloudflare Workers / Durable Objects gateway, with a Next.js frontend.
JVM Concurrency Benchmarks (open source)
JMH microbenchmarks for low-latency Java 21, with measured results. A shared counter under 1, 4 and 8 threads (LongAdder about 19x faster than AtomicLong at 8 threads), a hand-written lock-free SPSC ring buffer at roughly 3x the best JDK queue and 7x to 9x the blocking ones, and the cost of one allocation per event under G1, Parallel and ZGC (40 bytes and about 5x of the time, gone when the event is reused). It also records two benchmarking traps hit along the way: escape analysis hiding allocation, and timer-dominated percentiles at nanosecond scale.
Skills
- Languages
- Java, Go, Python, SQL, JavaScript
- ML / AI infrastructure
- Model deployment & serving, model optimization (quantization), inference latency/cost optimization, multi-vendor LLM orchestration, LiteLLM Proxy, RAG & vector search (Weaviate), LangGraph (multi-agent DAGs), LangChain, DSPy, LLM evaluation & observability (Langfuse), RAGAS, prompt engineering, LLM guardrails, agentic AI, MCP (Model Context Protocol), Gemini / OpenAI / Claude APIs, Ollama
- Backend & distributed systems
- Microservices, Hexagonal Architecture (Ports & Adapters), event-driven design, concurrency (asyncio, multithreading), multi-tenancy, low-latency optimization, REST, gRPC, FastAPI, Spring Boot, Spring MVC, Spring Framework, Spring AOP, design patterns, OOD
- Messaging & data
- NATS JetStream, Apache Kafka, PostgreSQL (pgx, PgBouncer), Redis, Weaviate, MongoDB, Elasticsearch, DynamoDB, Apache KVRocks
- Cloud & DevOps
- AWS (EKS, Lambda, S3, EC2), Docker, Kubernetes, Terraform, OpenTelemetry, Prometheus, GitHub Actions CI/CD, Airflow, Git, Maven, Gradle, JIRA, JUnit
Writing
- What If Your CyberSecurity System Knew the Attack Was Coming? (Medium, Mar 2026). A 16-minute technical essay on behavioural threat-intelligence architecture: the shift from indicators of compromise (IoC) to indicators of behaviour (IoB).
- Contributor, OCA IoB Working Group: open event-schema standards for distributed AI platforms.
Education & Certifications
Vellore Institute of Technology (VIT)
- OCI Generative AI Professional (Oracle, 2025)
- AWS Certified Cloud Practitioner (2025)
- Java Development on Oracle Cloud, Oracle University (certificate of completion, Feb 2024)
Contact
Open to Backend Engineer, Software Engineer (SDE 1) and AI / ML Infrastructure Engineer roles in Bengaluru, London, New York, Amsterdam, Hong Kong, or remote. Notice period is 30 days and negotiable to 15.
- Email: anshs5103@gmail.com
- GitHub: github.com/AnshSaxena05
- LinkedIn: linkedin.com/in/ansh-saxena-1c
- Medium: medium.com/@anshs5103
- LeetCode: leetcode.com/u/AnshSaxena1 (1,701 rating, 368 problems, Top 13%)