Software Engineer (Junior/Mid) — Agentic AI (Quant Research Platform) | Chennai (on-site)

Jnaara; Chennai District; On-site; INR 1,500,000 - 3,000,000; Full time

Benefits offered by this job

Relocation support available

Equity-based grant

Clear path up: comp/scope re-benchmark

Job summary

Jnaara in Chennai, on-site, seeks an early-career engineer who codes exceptionally well and wants to grow fast inside our AI-native research system. You’ll work directly with our senior team, shipping to institutional users, and learning how research-grade AI systems are actually built.

You’ll contribute to backend and async systems, multi-agent orchestration, data pipelines, and observability, using Python, FastAPI, Redis, Celery, and AWS.

Qualifications

• 2-4 years building real software; internships counted.
• Public GitHub with a stellar agentic-AI project you built yourself.
• Outstanding coder with clean, tested, typed code.
• Strong Python and solid CS fundamentals.
• Hands-on experience with LLM/agentic systems in production or exposure to quant finance.
• Ability to explain and defend code without tool-assisted prompts.
• High ownership and fast iteration in a senior environment.
• Open-source contributions to agent-infrastructure or ML tooling.

Responsibilities

• Build agent workflows with inter-agent communication, tool delegation, retries, and error recovery.
• Implement context and memory components: persistence, retrieval, and reasoning traces.
• Develop async-first Python/FastAPI services handling long-running jobs.
• Handle task orchestration, caching (Redis), queues (Celery), and compute pipelines.
• Transform complex financial data into structured outputs and evaluable results.
• Contribute to observability: tracing, latency profiling, and instrumentation.

Skills

Python

CS fundamentals

Testing

Open-source

Agent orchestration

Problem solving

Tools

React

Next.js

TypeScript

AWS

Terraform

GitHub Actions

Redis

Celery

Job description

We're building an AI-native research system that emulates how top investors think - transforming complex data, ideas, and workflows into structured, decision-grade outputs. This is a systems + infrastructure problem, not a wrapper.

Jnaara is built by veteran researchers, portfolio managers, and CTOs from renowned hedge funds and asset management firms. We work closely with a $200B+ global asset management firm as a co-build partner - real workflows, real constraints, real users from day one.

This role is for an early-career engineer who codes exceptionally well and wants to grow fast inside that system - working directly with our senior team, shipping to institutional users, and learning how research-grade AI systems are actually built.

The Technical Challenge

Our platform runs many AI agents collaborating across complex, multi-step workflows - each with different tools, data access patterns, and reasoning strategies.

• Workflows are long-running, stateful, and non-deterministic
• Outputs must be reproducible, explainable, and auditable
• Systems must balance latency, cost, and reasoning quality

This is not prompt chaining. You'll be helping orchestrate intelligent systems under real-world constraints.

What You'll Work On

Multi-Agent Systems

• Build agent workflows: inter-agent communication, tool delegation, retries, and error recovery
• Implement context and memory components: state persistence, retrieval layers, reasoning traces

Backend & Async Systems

• Build async-first Python/FastAPI services handling concurrent workflows and long-running jobs
• Work with task orchestration, caching (Redis), queues (Celery), and compute pipelines

Data & Evaluation

• Build pipelines transforming complex, heterogeneous financial data into structured outputs
• Help build evaluation harnesses for output quality - golden datasets, regression tests, LLM-as-judge - so agents are measured, not vibes-checked

Observability

• Instrument tracing, latency profiling, and usage monitoring
• Make AI systems debuggable, inspectable, and auditable at every layer

Frontend (bonus, not core)

• Contribute to React/Next.js interfaces for inspecting workflows, comparing results, and streaming intermediate outputs

Frontend: React, Next.js, TypeScript

AI Layer: Multi-agent orchestration, retrieval systems, LLM APIs

Infra: AWS, Terraform, GitHub Actions

• 2-4 years building real software (production internships at strong companies count toward this)
• A public GitHub with at least one stellar agentic-AI project you built yourself - an agent harness, orchestration layer, eval framework, or memory system with real engineering behind it: original code (not forks or tutorials), tests, a README that explains your design decisions. This is a hard requirement - we open every repo, we read the code, and it's the first thing we'll ask you to defend live. Link it in your application; applications without it won't be reviewed.
• An outstanding coder - clean, tested, typed code you're proud to defend line by line
• Strong Python; solid CS fundamentals (we notice compilers, systems projects, and competitive programming)
• Hands-on experience with LLM/agentic systems in production, or genuine exposure to quant finance / trading / markets - either is a strong start, both is rare and we'll move fast
• You use AI coding tools fluently and can explain and defend every line without them - our process tests both, and we value full transparency about how you build
• High ownership, fast iteration, comfortable being the least experienced person in a very senior room
• Agent orchestration, eval pipelines, RAG, or memory systems you built yourself - side projects with real engineering count
• Interest in how investors think - markets, backtests, research workflows
• Open-source contributions to agent-infrastructure or ML tooling projects
• A compiler, systems project, or hardware/embedded tinkering habit
• Startup exposure or anything shipped 0-1, at any scale

Chennai, on-site. We're a small team building fast, in person. Relocation support available.

• ₹15-30 LPA, calibrated to demonstrated level - where you land in the band depends on your take-home and live defense, not your years
• Equity: performance-based grant, formally reviewed at the end of your first year - we'd rather size it to demonstrated impact than guess on day one
• Clear path up: comp and scope are re-benchmarked as you prove out, not renegotiated from scratch

Our Process

Short intro call - a take-home you'll genuinely enjoy - a live session where you defend your submission, and your GitHub project, with our senior engineers. We move in days, not months.

Why This Is Different

Most AI startups wrap APIs, optimize prompts, ship demos.

We're building a research engine - with real institutional users, solving high- stakes problems - where systems thinking beats prompt engineering, and where a 2-year engineer who codes brilliantly gets responsibility most companies reserve for year eight.

If you care about building systems that think, not just respond, we should talk.