Full-Stack Backend Engineer

Company Description

COSO is a construction intelligence company building vertical AI for the construction project management industry. We embed AI agents inside multi-billion dollar infrastructure programmes. We are a growth-stage deeptech company with a working product already deployed on live infrastructure projects. Our team brings combined decades of experience across the infrastructure industry, enterprise software, AI, ML, and data science, spanning India and international markets.

Role Description

You will own the AI and data backbone of COSO end to end: ingestion, embeddings, retrieval, agentic orchestration, and model serving. You will join a lean AI engineering team, shipping into production every week. This is a builder seat, not a research seat. You will design the systems, write the code, debug the prod incident, and watch real users use what you built.

What you will do

• Work directly with the Head of AI Engineering on architecture, and with the founder on customer-facing deployments

• Build ingestion pipelines for messy, multi-format construction data

• Own the embeddings and retrieval infrastructure

• Build the scheduling intelligence engine

• Build agentic orchestration on the Anthropic SDK

• Stand up the evaluation and observability layer: golden sets, regression tests on prompts and retrieval, latency and cost dashboards, traces for every agent run.

• Design knowledge-retention strategies so the patterns we learn from customer data are durable and contractually defensible

• Ship to production with Docker, CI/CD, and clean deployment hygiene on cloud, with Postgres, Redis, object storage, and a queue.

• Enforce citation-verified outputs

Qualifications

•Computer Science background or equivalent practical depth.

• 2 to 5 years of post-qualification experience in backend engineering, production AI / ML systems, data engineering, or applied data science, with at least one production system you owned end to end and can walk us through in detail.

• Strong Python with async patterns, FastAPI, Pydantic, and clean typed code.

• Hands-on experience with PostgreSQL, Redis, object storage, and a task queue.

• Real production exposure to LLM systems: Anthropic and OpenAI SDKs, prompt engineering, function calling, and ideally at least one retrieval-heavy product.

• Working knowledge of vector databases and modern retrieval techniques: hybrid search, reranking, query rewriting

• Statistical literacy is important, you can fit distributions to messy real-world data, sanity-check the result, and not confuse correlation for causation.

• Comfort with Docker and shipping containers to production with CI/CD.

• Clear thinking on evaluation: you have built golden sets, run offline evals, and instrumented LLM apps in production.

• A track record of owning systems end to end, from spec to deploy to on-call.

Bonus points

• Numerical computing or simulation work shipped in production.

• Fine-tuning experience (LoRA, QLoRA) and ML ops tooling.

• Kubernetes, Terraform, or other infrastructure-as-code experience.

• GraphRAG, knowledge graphs, or structured retrieval at scale.

• IFC parsing, GIS data, BIM, AECO, or any prior construction tech exposure.

• Multilingual NLP including Hindi, Indic OCR or ASR, document AI (LayoutLM, Donut, ColPali).

• Distributed systems instincts and a willingness to debug at the network and storage layer.

• Heavy daily use of Claude Code, Cursor, or other AI coding agents in your real workflow.

Personality and mindset

• Hustler and hard worker. You enjoy shipping over theorising, and you measure yourself by what is in production.

• Hands on with deployment. You are equally comfortable designing the retrieval logic and debugging it when it breaks in production.

• End-to-end ownership with Head of AI Engineering. You scope, build, ship, monitor, and iterate quick in a lean team.

• Growth mindset. Flexible, collaborative, and quick to learn from feedback, the team, and customers.

What you will get

• True ownership of the AI backbone of a vertical AI company, with architectural authority on day one.

• Meaningful equity participation for the right candidate. Final compensation structure can be discussed based on candidate needs as well.

• Exposure to a real, working vertical AI product deployed with real customers running real infrastructure programmes.

• An AI-first internal stack with heavy use of Claude Code, Cursor, and Anthropic agents in our own workflow.

• A steep, compounding learning curve in agentic systems, retrieval, and applied AI on a frontier domain.