WhyPred

We build AI for investors

WhyPred is an AI consultancy for funds and professional investors. We design and build research workflows and applications tailored to your process.

Book a callSee an example build

Every call comes with its why

An investment committee will not act on a black box. So the systems we build read the way an analyst does, then hand you the call and the reasons, each tied to a source you can open. Here is an example: a morning brief agent.

  • Reads everything new against your book and your mandate
  • Writes the call and the reasons behind it
  • Waits for a person before anything leaves the building
Morning brief agentExample build, fictional companies
Read4 of 4

38 holdings, 14 announcements since the close, 2 need you

  1. Kestrel Point Minerals: quarterly activities report18 pages, lodged 06:12
  2. Kestrel Point Minerals: investor presentation24 pages, lodged 06:14
  3. Your book: positions and mandate limitsAs at last night’s close
  4. 11 other announcementsRead in full, nothing material
Call

Review Kestrel Point Minerals before the open

Why
  • Mill grade at the main deposit ran 18% below the feasibility plan, for a second quarter running.1
  • Cost guidance is unchanged, but year-to-date all-in costs already sit above the top of the range.1
  • The presentation drops the 2027 expansion timeline that was on slide 6 last quarter.2
  • At 3.1% of NAV the position is over your 2.5% soft limit for single-asset miners.3
Confidence: MediumTo: PM, Risk

What we build

Not a platform to adopt. Systems engineered around how your desk works, then handed over so the edge stays with your team.

AI agents

Agents that watch your universe while you sleep, read what is new and flag what matters to your book. A person approves anything that leaves the building.

  • Announcements, filings, transcripts, news
  • Your positions and mandate rules as context
  • Every action logged
Approval neededWaiting on you

Morning brief agent wants to send

Kestrel Point: grade miss, position over soft limit

To
Portfolio manager, Risk
Cites
Quarterly report, investor presentation, your book
  1. 06:10Read 14 announcements across 38 holdings
  2. 06:13Flagged 2 against your mandate limits
  3. 06:15Drafted a brief with 4 cited reasons

Research workflows

The repeatable work behind every reporting season, rebuilt as a pipeline you can run, check and audit: data in, decision-ready notes out.

  • Built on your data and your model
  • Differences surfaced, never smoothed over
  • Runs on a schedule or on demand
Quarterly production reviewStarted 06:00
  1. Pull quarterly reports for 12 miners1m 48s
  2. Extract production, grade and cost lines3m 10s
  3. Reconcile against your model3 differences0m 42s
  4. Draft variance notes1m 05s
  5. Analyst reviewWaiting

Frontier and private models

The strongest model for each task. Claude and other frontier models where that is allowed; open-weight models on your own hardware where your data must stay home.

  • Chosen per task, not per vendor
  • Client and portfolio data stays where your mandate says
Model routingSet per task
TaskRuns on
Summarise public filingsClaudeEnterprise API
Read client holdingsOpen-weight modelYour servers
Draft client lettersClaudeEnterprise API
Score trade ideasYour own quant modelYour servers
~/research/comps $ claude
> Add the four new lithium producers to the comps
  pipeline and rerun the tests

  Read pipeline/comps.py, config/universe.yaml
  Added 4 companies to universe.yaml
  Mapped spodumene grades to the shared unit (Li2O %)
  Ran 38 tests: 38 passed

> _

Claude Code training

Hands-on training that gets analysts and quants building their own tools with Claude Code, in your codebase, on your problems. Then we build alongside them.

  • Workshops for investment teams
  • Implementation in your repositories

Quant research and validation

Signal research, model and manager validation, and backtests held to the standard capital needs: out of sample, after costs, and written up for review.

  • Independent review of models you already run
  • Plain verdicts, including “no”
Validation: earnings revision momentum
MeasureBacktestOut of sample
Sharpe ratio1.420.38
Annual return after costs9.8%1.1%
Annual turnover610%640%
Worst drawdown−12.4%−19.7%
VerdictMost of the edge is in-sample. Not ready for capital.

How an engagement runs

Working software at every step, and nothing at the end that only we can run. You decide how far it goes after the first system is live.

  1. Scope

    We sit with your desk, map the process, and name where AI moves the result and where it does not.

  2. Build

    We build against your data and workflow in short cycles, with your team reviewing each one.

  3. Hand over

    You keep the code, the prompts, the models and the documentation. No platform licence, no lock-in.

  4. Train

    Your people learn to run, check and extend it, with Claude Code training if they want to build more.

Markets first, then code

WhyPred is a Sydney practice run by people who have sat at the desk. We build what we would have wanted there.

Over a decade in investment management
Paired with hands-on software and AI engineering, so the systems are built by someone who knows the work.
Teaches AI and quantitative methods
At the University of Sydney, UNSW and Kaplan Business School.
Peer reviewer for IEEE journals
Holding machine learning work to a reviewer’s standard, ours included.
Based in Sydney
Meet in person in Sydney, or on a call wherever you are.

Book a call

Thirty minutes on where AI can sharpen your process, and where it cannot. Bring one workflow you would like to stop doing by hand.

Office
Sydney, Australia
Pick a time30 minutes, by video