What's In It For You

- A core engineering seat: the agents are the product, and they are yours.
- An anchor client engagement already underway: your work ships into real, paying work within weeks, not a speculative roadmap.
- Competitive compensation.
- A genuinely AI-native engineering culture: agentic coding tools from day one, agent-per-human leverage, and the friction you hit becomes product input: we are user zero of our own platform.
- No legacy codebase and no bureaucracy, working alongside founders with deep services and product pedigree.

- The chance to build for regulated, high-stakes domains such as healthcare, life sciences, financial services, where quality and trust are the actual product.
- A modern office workplace in Gurgaon, with a clear path to grow as the engineering organisation scales.

Duties & Accountabilities

- Agent engineering: design and build the initial BA, Dev, and QA agents: prompt architecture, context assembly, tool use, and schema-validated JSON outputs, with prompts and skills packs versioned in git and treated as core IP.
- Retrieval & context: build the codebase retrieval layer (syntax-aware chunking with tree-sitter, embeddings in pgvector) and the context-pack system: project glossary, org conventions, constraint register, and exemplar retrieval.
- Evaluation harness: own the golden dataset (25–30 expert-verified items) and the harness that runs nightly in CI: per-agent scores against published acceptance bars, trend monitoring, and regression alerts that block release.
- Failure modes: design for the ways agents fail: schema-repair loops, reject-loop re-invocation with reviewer comments injected, retrieval-verified symbols to guard against hallucinated APIs,



and honest gap declaration over coverage inflation.
- Confidence & quality: implement and tune the confidence-scoring model that routes low-confidence outputs to senior review, and use every human modification (deltas captured at the gates) as improvement signal.
- Token economics: own cost per story: per-invocation token and cost telemetry, model-mix tuning (fast tier for enrichment and test generation, top tier for code), and budget-aware prompt design.
- AI-native practice: work daily with agentic coding tools; the friction you hit is product input.

What We Will Look For

- You ask how acceptance is measured before you ask what the prompt looks like. An evals-first mindset is the defining trait of this role.
- You have a working vocabulary for failure modes (hallucination, drift, coverage inflation, silent assumption-making), and you design around them rather than explaining them away.
- You treat prompt changes like code changes: versioned, tested against a regression set, and rolled back when the numbers say so.

Education & Experience

- 3+ years of software engineering experience, including 1.5+ years building LLM-powered applications in real production use, not prototypes or side projects.
- Demonstrable experience with retrieval-augmented generation and context engineering in a production system.




- Experience building or operating systematic evaluation for LLM outputs (golden datasets, regression harnesses, rubric scoring).
- Experience in a startup or small-team environment is a strong plus.
- Bachelor's degree in Computer Science, Engineering, or a related field, or equivalent demonstrable experience.

Technical Skills Required

- Strong Python and production experience with LLM provider APIs (Anthropic Claude family experience valued); structured outputs and JSON-schema validation.
- Prompt and context engineering with measurable iteration, exemplar selection, context budgeting, instruction design for consistency.
- Embeddings and vector search in production (pgvector or similar); code-aware chunking and indexing is a strong plus.
- Evaluation engineering: building harnesses, defining acceptance bars, wiring them into CI, and reading the trends.
- Token and cost optimisation: metering, model-tier routing, and cost-per-outcome thinking.
- Agent orchestration frameworks (LangGraph or similar) are a strong plus, as is SQL fluency for telemetry analysis.
- Practical, hands-on fluency with modern agentic coding tools (e.g., Claude Code or equivalent) is required.

How We Hire Our process is designed to move from first contact to offer in about a week: a 30-minute screen (including a short live demonstration of how you work with agentic coding tools such as Claude Code; every engineer here works this way), a paid four-hour work sample, a founder interview, and references. If we are a mutual fit, you will know fast.

To apply, or to discuss the role directly, reach out to the team.

📌 Artificial Intelligence Engineer (Gurugram)
🏢 GoodSpace
📍 Gurugram