Richard Hermann

Over 30 years of alternative investment experience, with a track record of identifying overlooked opportunities.

Finding what others missed in investments, operational risk, strategy, and software.

40%
Past 12 monthsNet return on a highly diversified hedge fund I recommended, with near zero net market/sector exposure, as of August 30, 2026.1
1/100th
CurrentThe cost of LLM extraction, from a patent-pending method for parsing alternative investment statements, with material improvements over LLM accuracy and speed in addition to cost.2
85%
CurrentError reduction in operational processes at one firm and a savings of $4 million per year at another.2
  1. Approximate, and net of fees. These are results of investments I recommended. Past results do not predict future results.
  2. Approximate.
Investments
Alternative investment research, marketing, and risk management. Experience with long/short equity, market neutral, long/short credit, high yield, distressed bonds, distressed muni-bonds, structured credit and CMOs, Brady bonds, real estate (residential and commercial - parking), emerging markets debt, private equity
Software
LLM accuracy through governed, permissioned context and deterministic methods that avoid hallucination; text-based risk management for financial research; operational risk reduction through feedback on human-run processes
Solutions
Operational due diligence, position-level risk monitoring, fund administration and investor reporting, error reduction in regulated manufacturing, analyst training programs. Several were the industry’s first.

For more than thirty years I have worked in senior positions across nearly every part of the alternative investment business covering much more than the typical allocator or marketer: fund research, risk management, manager due diligence, legal, compliance, operations, fund formation, software development and capital raising. Along the way I launched onshore and offshore hedge funds and raised more than $500 million for early-stage managers.

I am also a full-stack developer who sold my first software before I could drive. When the tool an opportunity needed didn’t exist, I built it, and several were firsts in the industry.

Sometimes it is the investment that was ignored. Sometimes it is the software nobody has written yet. Sometimes it is a solution never envisioned. Sometimes it is the risk that was never recognized.

Selected results

PeriodResultBasis
2025–2640% net return1A diversified equity hedge fund I recommended in August 2025, over the twelve months that followed, with nearly zero market and sector exposure.
Current1/100th the cost of LLM extraction2A patent-pending method for parsing semi-structured alternative investment statements, using non-text vectors for document alignment, at 99.7% accuracy. In production in AltsByClick.
Current85% fewer errors in QC testing; manufacturing errors lowered by $4 million per year2The error reduction was measured with operator guidance software in QC testing at a biopharma. The $4 million is the yearly cost of manufacturing errors removed at a large medical technology manufacturer that adopted an operations process recommendation of mine.
2007–0960 points ahead of the S&P 5001Six hedge fund investments I recommended to a family office, moving it away from credit risk before the financial crisis. More than 20% net over 18 months while the index fell roughly 40%.
2007Capital preserved in fullRecommended that a family office redeem from the Bear Stearns Structured Credit Fund. The fund later collapsed into bankruptcy.
  1. Approximate, and net of fees. These are results of investments I recommended. Past results do not predict future results.
  2. Approximate.

Opportunities uncovered

The calls and the firsts, and how each came about.

  • Prescient credit call ahead of the crisis

    2007 to 2009

    In 2007 I urged a family office to diversify away from its portfolio’s credit exposure and, at the same time, recommended six hedge funds that were positioned well for a more risky market environment. Over the next 18 months they returned more than 20% net while the S&P 500 fell roughly 40%.

    Eighteen months from 2007

    Six hedge fund investments, net of feesover +20%

    S&P 500, same 18 monthsabout −40%

  • Three patents pending, with the code written by the inventor

    Portside Group, 2003 to present

    Designed the methods, filed the patents, and write the production systems myself: more than 10,000 lines of Python behind live deployments in financial services and biopharma, with full-stack delivery on AWS in React, TypeScript, and infrastructure as code. On alternative investment statements, the parsing runs at 99.7% accuracy.2

  • Kept a family office out of a fraud

    Portside Group

    Independent due diligence led me to recommend against a manager a family office was preparing to hire. The manager was later shown to be fraudulent.

  • A mispriced synthetic futures contract

    Portside Group

    I caught an error in the pricing of a synthetic futures contract that Goldman Sachs, our portfolio manager, and the trading department had not caught.

  • The first back-office audit of hedge funds

    Collins Associates, 1990s

    When investors judged hedge funds almost entirely on strategy analysis and returns, I created a structured audit of a manager’s operations, service providers, technology, and internal controls, and trained a team to apply it to every manager. Operational due diligence has since become standard practice across the industry.

  • The first software to read a prime broker’s read-only reports

    Collins Associates, 1990s

    Prime brokers exposed convertible bond arbitrage positions only as read-only mainframe reports for internal risk monitoring. I wrote the software that parsed them, which gave our firm independent, position-level risk monitoring of managers with whom we invested, and then used the broker’s own analytics to identify the hedge behind each position.

  • Launched onshore and offshore hedge funds, and wrote the accounting software for self-administration of one

    Fuller & Thaler Asset Management, 1999 to 2003

    As partner and Director of Operations at a $2 billion institutional asset manager, I handled legal, SEC compliance, reporting, trade reconciliation, and receivables, and was responsible for fund formation for both onshore and offshore hedge funds. I also wrote the partnership accounting software that tracked general-partner allocations and investor performance. Separately I envisioned, architected, and helped write the firmwide client reporting system that cut the monthly client reporting cycle from two weeks to thirty minutes.

AI research

My research focuses on improving the accuracy of LLM processing through the use of non-text vectors, often orthogonal to the vectors LLMs rely on. The results are not only improvements in speed, but also accuracy and cost.

Deterministic alignment in place of LLM extraction

In production as AltsByClick, formerly DocEdge AI

Question
Capital account statements, capital calls, distributions, and valuations arrive as PDFs in whatever layout each administrator prefers. LLMs misread them in ways that are hard to detect. Can they be parsed without trusting an LLM to read them correctly?
Method

The patent-pending approach profiles every item on the page and uses Shannon entropy to rank which features actually tell one field from another. It identifies the landmarks, the labels that maintain relationships to each other, from one statement to the next, then globally aligns each new statement to the last one from the same source, so a value is found by where it sits relative to those landmarks.

The process utilizes a feedback loop for context improvement that leverages bi-directional and cross-source validation. As an improvement over RAG, the process uses Query-Directed Multi-Representation Evidence Retrieval.

Result
Accuracy does not rest on the model. Every match is validated in both directions, and reconciliation checks flag any figure that is inconsistent or unsupported. It is live with a family office, consolidating a multi-custodian portfolio that is mostly alternatives into one view, and it feeds a monitoring tool that computes beta, volatility, drawdown, and related risk analytics.
Side benefit
No templates and no manual entry, which is where the one-hundredth cost figure comes from.

Verifiable, governed context for LLMs

Question
In regulated environments, how can context be shared with trusted parties, within bounded use, and with privacy protection?
Method
A federated architecture versions, verifies, and dynamically assembles governed context for LLM applications, so that provenance and verification are infrastructure rather than afterthoughts. It grew out of the need for LLMs to safely utilize private context information for agentic workflows through a Model Context Protocol server.
Result
A working evaluation method decides which tasks a model is allowed to perform, what context it can access, and legally binds restrictions on use. This is all in patent-pending registry design, listed under Patents below.

Process errors by operators in regulated environments, reduced through systematic feedback

Question
This is risk management applied to operational processes in a regulated environment, here biopharma. Operator error is operational risk, the same kind I audited in hedge fund back offices. Can the methodology that keeps an AI on a checkable path do the same for an operator, preventing the error as the work is done rather than finding it afterward?
Method
The creative, thoughtful solution others had overlooked: give operators feedback on each run (with 30+ steps which can have confounding interactions, it is far from trivial, so computational methods are required for root-cause analysis), and score performance on how faithfully results follow the perfect path. Patent pending.
Result
The 85% reduction in the results table was measured in live trials at a San Diego contract manufacturer, across eight production runs of a 42-step potency procedure. The same thinking, delivered as a process recommendation rather than software, is behind the $4 million per year at a large medical technology manufacturer.
Side benefit
Operator execution is continually improving towards perfect.

Writing

Investors are systematically overpaying on alternative investment concentrated bets

AI investments, because of their volatility, could be one of the worst instances. A flat 20% carry is worth nearly twice as much on a concentrated bet versus a diversified fund due to increased volatility.

Working paper, first drafted December 2010, revised July 2026

An incentive fee is a call option, so it is worth more when the underlying is more volatile. The paper draws out the consequence that gets less attention: a single-name position is far more volatile than the diversified fund its terms are borrowed from, so the same twenty percent is worth close to twice as much on the isolated bet. I call that wedge the concentration premium. At the illustrated volatilities the position must out-earn the fund by roughly eighty basis points a year just to pay for its own fee structure, before any compensation for undiversified risk.

The paper argues the premium survives because both sides of the table sincerely believe the same wrong number, and proposes fee terms that hold up anyway: carry scaled to position volatility, cumulative measurement across investments, clawbacks, and high-water marks.

Richard Hermann, “Incentive Fees, Alignment of Interests, and Rational Decision-Making in the Context of Special Opportunity Investments.” 13 pages.

Read the paper (13 pages).

  • Point/Counterpoint: Eating One’s Own Cooking MAR Strategies, no. 13, July 1999, p. 10
  • Conflicts of Interest Regarding Hedge Funds Private Asset Management, April 5, 1999, p. 10
  • Letter from America: Risk Management. Hedge Fund Community Embraces Mathematical Approaches but Neglects Operational Risk AIMA Newsletter, September 1998, pp. 26–27
  • Portfolio Ponder: Why Hedge Funds Aren’t Always a Win-Win Interview by Sharon Frederick, Comstock’s Magazine, February 2009

Patents

Inventor on three U.S. patent applications, all patent pending.

  • Adaptive Document Content Extraction via Entropy-Guided Global Alignment US Patent Application No. 19/328,817 (Pub. US 2026/0080704 A1, Mar. 19 2026), pending. Shannon-entropy feature profiling, landmark identification, and global alignment with bi-directional validation. The core mechanism behind AltsByClick, with material improvements over LLM accuracy, speed and cost. Read the published application (PDF, 22 pages). Also at the USPTO and on Google Patents.
  • Binding Verifiable Context from Authoritative Sources to Digital Artifacts for Semantic Integrity and Federated Governance U.S. patent application, patent pending Automated, verified context assembly for LLMs. Each piece of context is bound to its authoritative source, so that before it is used, its origin can be authenticated, the user’s right to it can be checked, and the contractual terms of its use are carried with it.
  • System and Method for Systematic Human Mastery Development, Training, Assessment, Fidelity Scoring, and Allocation via Multivariate Perfect-Path Feedback and Sub-Task Alignment U.S. patent application, patent pending Trains and assesses people on complex procedures by scoring, sub-task by sub-task, how faithfully each run follows the perfect path. The scores feed back to operators automatically, so every run is corrected against perfect execution and each repetition moves closer to it.

Career

2003 – present
Portside Group, San Diego. Partner: alternative investments, risk management, capital raising, and software development, including AI.
1999 – 2003
Fuller & Thaler Asset Management, San Mateo. Partner and Director of Operations at the $2 billion behavioral finance manager where Nobel laureate Richard Thaler is a partner.
1993 – 1999
Collins Associates, Newport Beach. Hired for performance attribution at a $1.7 billion institutional fund-of-funds; promoted three times, to domestic research, international research, and Risk Manager.
Education
BA, Mathematics, University of California, Riverside, 1992. FINRA Series 7, 24, and 63. Co-founder of Bay Area Hedge Fund Professionals.

Roles, responsibilities, and the complete record are in the CV.

Speaking

Twenty-three conference talks between 1998 and 2012, on operational due diligence, risk management, and manager selection, from New York and Chicago to Bermuda and Hong Kong.

  • What a Back Office Says About a ManagerMAR Mid-Year Conference on Alternative Investment Strategies, Lake Tahoe, 1999
  • The Overlap Between Hedge Funds and Venture CapitalKeynote, Australian Consulate-General, San Francisco, 1999
  • Successfully Managing Risk in Alternative InvestmentsAIC Global Hedge Funds Congress, Hong Kong, 1999
  • Behavioral Mistakes Investors Should AvoidAmerican Association of Individual Investors, Silicon Valley Chapter, 2000
  • Risk Reporting to InvestorsManaged Funds Association Forum, Chicago, 2002
  • Identifying Hedge Fund Manager Alpha and BetaAlternative Asset Summit, Las Vegas, 2008

All twenty-three are listed in the CV.