Headline

Member of Technical Staff at Aegion AI | Backend & AI-infra engineer | LLM agents, RAG, MCP, vector DBs on AWS (ECS, SNS/SQS, Pulumi) | Python · Go · FastAPI · Kafka | Distributed systems at 300K-user scale

About

I own the whole vertical beneath an AI agent: the runtime, the retrieval, the integration bus, and the AWS infrastructure it runs on. At Aegion (Mira) that is a production customer-support agent on Agno, a RAG pipeline over Milvus, and a layer that folds 20+ platforms into one schema over SNS/SQS. The part I obsess over is what happens off the happy path. Agents fail quietly: a tool returns something slightly wrong and the agent keeps going. So I built passive detection that diffs tool input and output across runs and escalates to a human the moment a path goes dead. Escalate-first on writes, auto-retry only on reads, so a retry never double-fires a write. The through-line is ownership: taking a loosely defined problem front to back, and caring about the parts that make an AI system trustworthy: observability first, schemas over trust, boring tech on purpose. Before agents, I architected Mira's distributed LLM routing platform (FastAPI, Fastify, Go CLI) and scaled the volume it handled 5x, then built Klok, a crypto RAG app that grew to 300,000+ users. On the side I build things I want to exist: PennyWise, an Android expense tracker that parses bank SMS entirely on-device, no cloud (510 GitHub stars, 1,000+ installs), and Guardian, an AWS service-discovery platform with an AI doc chatbot. Stack: Python, Go, TypeScript; FastAPI, Agno, MCP, RAG; Milvus, Postgres, Redis; Kafka, AWS (ECS Fargate, RDS, Lambda), Pulumi; Langfuse and OpenTelemetry. github.com/sarim2000 · sarimahmed.net

Location

India

AI stack

antigravity

claude code

codex

cursor

Work experience

Member of Technical Staff at Aegion AI, Mar 2026 - Present · 6 mos, Bengaluru, Building Aegion (aegion.so): production AI agents for customer operations. Scope: the agent runtime, retrieval, integrations, and the AWS infra beneath them. - Rebuilt the production customer-support agent on Agno (off CrewAI): a 20+ tool registry, Milvus short and long-term memory, scheduled reminders, passive issue detection, and auto-escalation routing. - Own the RAG pipeline end to end: retriever (Agno plus MCP over Milvus), ingestion (Google Docs, Notion, files, plus a web crawler built before Firecrawl existed, semantic chunking, chunk-level dedup), a multi-stage classifier, and an eval service with its own datasets. - Built a config-driven MCP tool factory for agent orchestration, with async tool registration and usage tracking. - Built the integration bus that folds 20+ external platforms into one schema over AWS SNS/SQS: OAuth syncs, per-tenant isolation, webhook signature verification, dead-letter replay, and backoff retries. - Run the AWS infra as code on Pulumi: ECS Fargate, RDS Postgres, ElastiCache/Redis, SNS/SQS with DLQs, Milvus on EC2, IAM role-based auth; Langfuse and OpenTelemetry for tracing and logs; SOC2 support. Stack: Python, Go, Agno, MCP, RAG, Milvus, FastAPI, AWS (ECS Fargate, RDS, SNS/SQS), Pulumi, Langfuse, OpenTelemetry.

Senior Software Development Engineer at Mira, Jan 2023 - Present · 3 yrs 8 mos, Singapore, Built and scaled Mira's core platform: a distributed LLM router and a 300,000+ user product on it. - Architected the Mira Network, a distributed LLM routing platform (FastAPI router, Fastify service nodes, Go CLI); scaled request volume handled 5x. - Built the Mira SDK, a YAML-based flow generator, and a Go CLI for multi-step LLM flows, adopted by 100+ developers. - Built and scaled Klok, a crypto RAG app ("ChatGPT for crypto news"), to 300,000+ registered users and 1,000+ concurrent daily: FastAPI on AWS RDS/Fargate, Redis leaderboards and caching, Milvus semantic search, and a load-balanced async architecture. Owned the backend, built parts of the frontend. - Built the data and infrastructure layer: a Kafka stream, Lambda for RudderStack, an in-house ETL pipeline, OLTP analytics on Glue and Athena, AWS WAF, and full LGTM observability (Grafana, Prometheus, Loki, Thanos with S3 backup) on AWS CI/CD. - Tech-led a pod of junior engineers: set code-review and mentoring standards, reviewed PRs, interviewed engineering candidates, and ran on-call and incident response. Stack: Python, Go, FastAPI, Fastify, Milvus, Redis, Kafka, AWS (RDS, Fargate, Lambda, Glue, Athena, WAF), Grafana/Prometheus/Loki/Thanos.

Information Technology Intern at ListenUp, Jan 2022 - Jul 2022 · 7 mos, 1. Improved database schema design in mongodb so as to increase efficiency of querying and reduce data redundancy 2. Accomplished a rating system using React Native’s PanResponder API where a user can swipe or touch to rate a song, it keeps saving until a user has finally decided on a rating and also giving ability to cancel at any time 3. Designed an artist tagging feature where an artists can credit another artist and it will link profile of respective artist while also checking whether tagged artist is an artist or not 4. Performed a backend functionality where a check is kept whenever a subscription has ended, thereby alerting user to either subscribe again or to cancel

Full Stack Engineer at Stader Labs, Sep 2021 - Jan 2022 · 5 mos, 1. Analyzed and developed web application using NextJS with team of 5 UI developers; interacted with 2 Senior Project Managers to gather requirements for front-end development tasks 2. Designed and developed a website so as to introduce the Stader team and also launch dates of various products of StaderLabs

Full Stack Engineer at Fashioo, Jun 2021 - Sep 2021 · 4 mos, 1. Worked on improving html, css and JS codes, maintained and reviewed it. 2. Created a video chat app 3. Learned about WebRTC and how to integrate it with nodeJS, create audio chat,call and session for users.

Education

Bachelor of Engineering - BE, Computer Science, Visvesvaraya Technological University

Public links

github: https://github.com/sarim2000

sarimahmed.net: https://sarimahmed.net/