INDIA (Remote) - Full-Stack Engineer (Pacific time)

About

About e3 group

e3 Group is a San Francisco AI company rebuilding freight brokerage around autonomous agents. We put voice and email negotiation, load matching, document processing, fraud detection, and legacy-system modernization into production for real brokerages — the unglamorous, high-stakes workflows that move freight and money every day. We're small, founder-led, and shipping fast: our agents do real work in front of paying customers, where a wrong number or a dropped call has consequences.

About the role

We're hiring full-stack engineers who can build production AI end-to-end and own it. This isn't a feature factory — you'll take a problem from the first customer call through the product decision, the code, the deploy, and the traces you watch afterward, then do it again the next week. The surface is real full-stack (React, Next.js, Python), but the center of gravity is applied AI: LLM-driven voice and email agents, streaming speech pipelines, tool-calling with hard guardrails, retrieval, and the evals that keep any of it trustworthy. It's a high bar, and we'd rather be upfront about how we actually evaluate people — because it's exactly how we run interviews: We go straight at the biggest claim on your resume. Tell us what you personally built, how it works under the hood, how you measured whether it was right, what broke, and what you changed. "My team did that" is fine — as long as you're clear about your part. Evaluation is non-negotiable. If your answer to "how did you know this was correct before any user told you?" is spot-checks or a good prompt, that's a flag. Golden sets, regression and hallucination checks, and tool-call correctness are the language we want. Voice and agents are the domain: streaming, barge-in, latency budgets, and deterministic checks around anything an LLM can get wrong — prices, terms, actions. Think from first principles. Reason about scale, cost, and context limits from scratch; reach for embeddings, retrieval, and reranking before brute force. Work like an AI-native engineer. Claude Code and Codex are daily tools, and you can verify what they produce instead of trusting it blindly. We're not chasing perfect. Some of our best hires needed a nudge on hard problems — what mattered was getting to the right answer, reasoning honestly, and owning their thinking.

What you'll own

• Own features end-to-end and ship them weekly: scope the problem with a customer, build across the stack, deploy, and watch the production traces yourself.
• Build the voice and email negotiation agents — the speech pipeline (STT, TTS, turn-taking), the LLM orchestration, and the deterministic rules that keep prices and terms correct.
• Build the evaluation layer: golden datasets, offline and online checks, regression and hallucination testing, and tool-call correctness for everything the agents do.
• Build retrieval and matching on embeddings and reranking, tuned for cost and latency at real scale.
• Instrument production observability for three audiences at once: engineers debugging, product verifying correctness, and customers who need to trust the outcome.
• Bring product judgment — tell the difference between a one-off customer request and a real signal worth building around.

Requirements

Must-have

• 5+ years shipping production software with real end-to-end ownership across a full stack (Python and TypeScript/React).
• Hands-on experience taking LLM features to production — architecture, tool-calling, and retrieval, not just prompting.
• Fluency with evaluation and testing for AI systems: how you know something works before a user tells you it doesn't.
• Strong system-design fundamentals — streaming, concurrency, SQL, and reasoning about scale and cost from first principles.
• Comfort with ambiguity and high-stakes calls on limited data, plus generative AI tooling (Claude Code, Codex) as daily instruments you direct and verify.

Nice-to-have

• Voice or real-time systems experience: STT/TTS, barge-in, latency budgets, telephony.
• Founding or early-stage engineer background — evidence you can operate without a playbook.
• Functional languages (OCaml, Haskell, Scala, Rust) or hardware description languages as a signal of depth and range.
• Multi-agent orchestration, negotiation flows, or other complex tool-use systems.
• Open-source work or personal projects shipping at scale; competitive programming as a supplementary signal.

Benefits & perks

• Fully remote work environment
• Competitive compensation
• Opportunity to work with a fast-moving engineering team
• High ownership and impact on product development

Interview process

1. 1Application review
2. 2Initial Screen - 15 min
3. 3Technical interview 1 - Live Coding
4. 4Technical Interview 2 - System Design
5. 5Interview
6. 6Offer Extended
7. 7Hired