Short answer
Lucent is an agentic AI platform for institutional investment research, built for wealth management, public equity and private-markets teams. It runs several frontier language models through purpose-built research agents, reconciles their outputs with citations and confidence scores, and produces editable research including valuation models and scenarios. It covers NZX and ASX companies and is a Levy Street venture.
- Multiple models are run and reconciled rather than trusting one model’s answer.
- Every output is grounded in citations an analyst can check.
- Coverage is live for NZX and ASX, and for private companies globally.
- LucentBench, described in its whitepaper, tests forward-looking earnings forecasts.
- What
- Agentic investment research platform for institutional teams
- Coverage
- NZX and ASX live; private companies globally
- Approach
- Several frontier models run together, reconciled with citations and confidence scores
- Outputs
- Research, DCF, comparables, scenarios and alerts, exportable to firm templates
- Structure
- A Levy Street venture, trading through Lucent Research Limited
What Lucent does
Investment teams are asked to cover more companies, react faster and meet stricter governance standards. The information they need is scattered across filings, results, transcripts, news, data rooms and internal notes. Lucent is built to pull that together and produce research an analyst can stand behind.
Its workflow runs in four stages, in Lucent’s terms: ingest, synthesise, analyse and act. Purpose-built research agents gather and read the sources; several frontier models analyse them independently; Lucent’s aggregation layer reconciles their views with citations and confidence scores; and the output becomes editable research, from summaries to discounted cash flow models, comparables, scenarios and alerts, exportable to a firm’s own templates.
Why it is built this way
- Model consortium
- No single model is trusted alone. Disagreement between models is surfaced, not averaged away.
- Citations first
- Every claim links to its source, so research can be checked and defended.
- Editable outputs
- Analysts change assumptions and the models update; the AI drafts, people decide.
- Governance
- Lucent lists enterprise single sign-on, role-based access, encryption in transit and at rest, audit logs and private deployment options.
These are the same principles we use in retrieval-augmented generation projects for clients: ground answers in sources, make uncertainty visible, and keep a human accountable for decisions.
LucentBench
Evaluating financial AI is hard because models may have seen historical answers in their training data. Lucent’s August 2025 whitepaper describes LucentBench, a forward-looking benchmark based on quarterly earnings forecasting across 100 NASDAQ and NYSE stocks, evaluated on a period the whitepaper says falls outside the models’ training data to reduce contamination. See LucentBench for the method and its limits, and our guide to AI agent evaluations for the general approach.
What it shows about AI in regulated work
Financial services is a demanding test of production AI: outputs must be accurate, traceable and auditable, and the cost of a confident mistake is high. Lucent’s design (several models, citations, editable outputs and audit trails) is the pattern we recommend for any high-stakes research workflow. Our guide to AI for New Zealand financial services covers the regulatory context.
Written by the Levy Street team in Wellington. Last updated 19 October 2026.