Fermi is an early-stage AI tutoring startup, an initiative of Meraki Labs, dedicated to revolutionizing education globally. Our mission is to democratize world-class tutoring by making it accessible to every student through advanced AI. We are a dynamic and lean team of educators, designers, and engineers, united by a strong conviction that thoughtful design and cutting-edge AI can profoundly transform the learning experience.
Led by experienced founders, including Mukesh Bansal (Founder – Myntra, CureFit, Nurix) and Peeyush Ranjan (VP-Google, CTO-Flipkart, Airbnb), we are building a product that aims to reshape how millions of students learn—turning academic challenges into accomplishments. We operate out of Bangalore, India, fostering a high-velocity environment that prioritizes impact, ownership, and direct collaboration to define and build the future of AI in education.
We’re hiring an AI Engineer who is a problem-solver first: someone who can ship, debug, and iterate fast. You’ll help build the core AI workflows powering our tutoring product—especially agentic tutoring systems—and harden them into production-grade, scalable, observable systems.
This role is intentionally not for everyone. If you want tight scope, predictable tasks, or “only research / only backend,” this won’t fit. If you like building real systems end-to-end and seeing your work hit production quickly, you’ll love it.
Build and ship core AI workflows for our tutoring app:
agentic tutoring flows (hinting, step-by-step guidance, misconception detection)
answer evaluation / grading and feedback loops
retrieval + grounding (content ingestion, chunking, embedding, re-ranking)
personalization (student memory, progress signals, difficulty adaptation)
multimodal pipelines (images/diagrams; bonus if you’ve touched voice)
Turn prototypes into robust production systems:
latency + cost optimization (caching, batching, streaming, fallbacks)
eval-driven iteration (offline test sets, regression checks, quality gates)
observability (traces, logs, metrics, prompt/version tracking)
reliability + safety (guardrails, refusal behavior, policy/age-appropriate output)
Own features end-to-end:
Collaborate closely with product/design/education to translate learning goals into AI behavior.
1 to 5 years of strong Python fundamentals; you can write clean code and also “hack” when needed.
Comfortable building services with FastAPI/Flask, writing APIs, and debugging production issues.
Practical understanding of Docker, local dev workflows, and basic deployment concepts.
Good CS foundations (data structures, basic systems thinking, debugging).
Hands-on experience with at least one:
You understand the difference between:
You’ve built something real:
internships, substantial course projects, shipped side projects, research engineering, or open-source contributions.
CS background preferred (or equivalent demonstrated capability).
Strong signal from top engineering programs in India or an equivalent track record (exceptional projects / OSS / internships that clearly show you can perform).
You’ll do well here if you:
default to ownership: you don’t wait to be told every step
can move fast without being sloppy
enjoy ambiguity and turn it into clear execution
treat quality as an engineering problem: tests, evals, instrumentation, iteration
Build core AI product (not demos) that students use daily.
Massive scope for learning: agents, evals, multimodal tutoring, production hardening.
Small team, high trust, direct impact—your code ships and matters.
Work with a team that cares about craft, not just hype.
You want remote/hybrid (this is onsite Bangalore only).
You prefer narrow tickets and minimal ambiguity.
You’re not excited about production engineering (monitoring, reliability, cost, latency).
You’re mainly looking for a “research-only” role.
