? Founding Engineer

Tempra Capital · Remote, US / India · Equity-only · Full-time

Tempra Capital is a Multi-Solution AI OS for founder-led independent RIAs managing $100M–$2B in assets. These firms often run with four or five people, carry the same regulatory obligations as firms fifty times their size, and spend far too much of their week on administrative work instead of advising clients. We’re here to change that.

We reduce that load without asking firms to compromise on compliance: every output that becomes part of a client record is reproducible and auditable, while everything else is generative. Holding that line cleanly is the hardest engineering problem in the product—and the reason the product exists. We’re built on the TempoDyn platform architecture, and our first cohort of ten founding firms is being signed now.

Why this role exists

Right now, one person owns architecture, security, product, and most of the build. That’s not sustainable. We’re not pretending it is. We’re pre-revenue, we pay in equity, and we’re looking for one engineer to own the platform alongside the founder who architected it.

You’re the fix: If you understand capital markets and can build agent systems, you’re a rare combination—and this role was written for you.There’s no backlog waiting for you. No ticket queue. No product manager handing you tasks. You’ll own the platform: what we build, how we build it, and what we refuse to build.

? What you’ll actually build

Agent systems

We run a portfolio of agents across:

- Investment research
- Portfolio construction
- Client communication
- Practice operations

That includes:
- Equity price forecast agents across 90-day, 180-day, and one-year horizons
- A buy/sell desk memo agent that consumes them
- A portfolio construction agent with a hard determinism boundary
- A constraint semantics layer
- Blocking validation rules

The determinism boundary

This is a big one. Some outputs must be identical every time they’re generated. Others benefit from a language model. Knowing where that line belongs—and building the architecture that enforces it—is a recurring design problem you’ll own.

Retrieval

We have two RAG corpora using hybrid dense + BM25 retrieval, tuned separately for advisor-facing and client-facing use. A safety classifier sits upstream of similarity search. Recall quality is a product feature here, not an infrastructure detail.

Integrations

You’ll work with real financial systems and real financial data:

- Wealthbox + Redtail → CRM
- eMoney + MoneyGuide → Planning




- Schwab + Fidelity → Custody

Trust infrastructure

You’ll work on

- Audit trails
- Reproducibility guarantees
- Compliance infrastructure
- The engineering work behind SOC 2 Type II

? Who we’re looking for

We’re not looking for a generalist who wants to learn finance on the job. And we’re not looking for a quant who has never shipped production software. This role sits at the intersection of capital markets + portfolio optimization + AI + production engineering.

? You understand capital markets

Not at the retail-investor level, but where you could sit in an investment committee meeting and follow every word. You understand:

- Asset classes and market structure: equities, fixed income, ETFs and mutual funds, alternatives—how they trade, settle, and what breaks.
- Fixed-income fundamentals: duration, convexity, credit spreads, the yield curve, and why a bond ladder is not the same product as a bond fund.
- Benchmarks, factor exposures, tracking error, attribution—and what a Sharpe ratio does and doesn’t tell you.
- Corporate actions, dividends, cost basis, wash sales, and the tax mechanics that can turn a technically optimal trade into a bad idea for a real client.
- The macro layer: rates, the FOMC calendar, earnings cycles, and how a forecast survives—or fails—contact with them.

? You can build agents for portfolio optimization

You understand

- Mean-variance optimization—and where it fails: estimation error in expected returns, unstable covariance matrices, and corner solutions.
- Practical remedies: covariance shrinkage, Black-Litterman, risk parity, resampling, robust optimization.
- Real advisor constraints: position limits, sector and concentration caps, minimum lot sizes, held-away and restricted positions, ESG screens, client-specific exclusions.
- Tax-aware construction: lot-level optimization, tax-loss harvesting, transition management for concentrated low-basis positions, turnover and transaction-cost penalties, drift-band rebalancing.
- When the answer is a solver—not an LLM.

⚙️ You build AI systems that survive production

You’re comfortable with

- LLM orchestration, tool schemas, multi-round tool loops, structured output extraction,



and clean agent handoffs.
- Evaluation as a first-class discipline: regression suites, adversarial cases, and detecting quality drift before customers do.
- Prompt architecture as engineering—with versioning, testing, and rollback—not copywriting.
- Strong Python.
- At least one major model provider, ideally several.

?️ You have compliance instincts

You don’t need to be a compliance skilled. But you should understand:

- Why RIAs care about books and records requirements
- Why a recommendation is a regulated act
- Why “the model said so” isn’t a defensible audit response

If you’ve built in a regulated industry before, that experience will pay off immediately.

? You have founder energy

You’re comfortable with

- Real product gaps
- An unwritten roadmap
- No cash compensation for a while

You’ll talk directly to advisors. You’ll call out what’s wrong. You’ll fix things without waiting for permission.

? The deal

We’re being direct because the alternative wastes everyone’s time.

Equity: 1%–3%, depending on experience and scope.

Vesting: Four years, one-year cliff, monthly thereafter.

Exercise window: Five years post-termination instead of the standard 90 days. You shouldn’t have to fund an exercise on a deadline to keep what you earned.

Cash: None right now.

Every team member is on the same terms, founder included. Cash compensation begins at our seed round or at sustained revenue, whichever comes first. Moving the team to market cash is a stated priority, not a vague promise.

? What this role is NOT

- Not a research position. We ship.
- Not an ML research role. We build systems on top of models rather than training them.
- Not a role for someone who needs a defined scope and a product manager.
- Not appropriate if you need salary in the next six months. That’s a reasonable need, and we’d rather say so now.

? How to apply

Skip the cover letter. Send us an email at [email protected] with your GitHub or a description of something you’ve built.

Then answer this one question in a few paragraphs:

A client holds $2M in a taxable account, of which $700,000 sits in a single stock with a cost basis of $90,000. The advisor wants them in a standard model portfolio. How would you design a system that produces a transition plan, and which parts of it should never be generated by a language model?

We’re not grading the finance. We’re looking at how you reason about the boundary between the two. Every applicant gets a reply.

📌 Founding Engineer (Bengaluru)
🏢 Tempra Capital
📍 Bengaluru