I build AI and web applications—and help turn ambitious ideas into ventures that can ship.
I'm Gregory O'Connor, an MIT-trained product builder, strategist, and co-founder of Socure. Through Vociferous AI, I work from opportunity and market analysis through architecture, implementation, evaluation, branding, and launch.
Current work spans public-source intelligence, creative tools for actors and filmmakers, AI coaching, research on the integrity of AI evaluation, and a stealth AI-adjacent venture for fields where completeness, accountability, and accuracy matter most.
Available for select consulting engagements: application development, product strategy, technical diligence, evaluation audits, and venture evaluation.
Selected work
Recent work across applications, venture building, and research. Each project below was taken from opportunity through a working system — public descriptions focus on what was built and decided; private methods and client information stay private.
- BNOW.NET — public-source intelligence, built and taken to market: a live, self-running intelligence product for analysts working fast-moving public information — continuous collection, a data-derived source-reliability registry, claims that stay attached to their evidence by database constraint, and a public scoreboard that grades the system's own output against expert analysis every day. I built the application and did the venture work around it: market analysis, pricing design, licensing strategy, and launch sequencing. Read more →
- SceneFiend.app — scene and monologue discovery for actors: a full consumer product — recommendations with stated reasoning, a curated metadata-only library, shareable piece pages, and a saved book actors rehearse from — built and operated by one person in six months on an AI-native workflow, and held in a small actor beta until its content-rights and quality gates clear. Read more →
- NaviaAI.com — advisory board, venture research, and product engineering: work on both sides of an AI coaching company — brand, market, and design research for the founding team, and hands-on engineering with the technical founder: clearing the technical debt that blocked the multi-project feature beta users were asking for, then shipping that feature across users' encrypted coaching records with a rehearsed migration and a rollback. Read more →
- Dreamlet.Studio — a curated home for female-gaze cinema: a live, filmmaker-first platform pointing young cinephiles to short films that show women as fully realized people. Co-founded with Jacqui de Jong, and built out from a four-hour sprint that placed 12th with a special mention at a NY Tech Week 2026 competition into a supply-first venture with film-by-film licensing, filmmaker agreements, and rights enforced in the product. Read more →
- Working Title: Epistemic — stealth AI-adjacent venture: developing systems intended to bring greater certainty to applications where completeness, accountability, and accuracy matter most. Potential vertical markets include pharmaceuticals, finance, law, and defense. The methodology remains private. Read more →
- Venture evaluation and advisory: business, brand, market, go-to-market, and financial analysis across fintech, infrastructure, legal and public-interest work, health, and emerging AI ventures.
I don't do political microtargeting, covert persuasion, or surveillance work. If your use case needs careful boundaries, we figure those out first.
Research
Before a score means anything, the answer key has to be right. Almost all evaluation effort goes into the model; almost none goes into the thing the model is graded against. My research audits that missing component — the benchmarks and answer keys that decide whether an AI system ships, gets bought, or gets trusted near a clinical or legal workflow.
- Two benchmark audits, under peer review: one of a widely used legal-reasoning benchmark, re-deriving every gold label from primary authority with the answer key sealed off; one of a public medication-safety benchmark whose construction rule infers that a drug is safe from the absence of a recorded risk. Titles, venue, and full findings stay private until the decisions land.
- A method built to be defended: the model stays off the measurement path, every item is checked rather than sampled, evidence grades are never summed into one flattering number, and everything that produced a result is attestable afterwards.
- Three engagements built on it: an Evaluation Audit, Gold-Standard Design, and Evidence & Provenance Design — for teams about to ship, buy, or defend an AI system on the strength of a score.
Where this fits
I'm most useful when a founder or team has a real opportunity, partial information, and pressure to decide and ship. I can help determine whether the opportunity should become software, build a testable first version, and connect technical choices to market, cost, brand, and operating reality.
- AI and web application development, from scoped prototype through release candidate and handoff.
- AI-native delivery: building a real product at unusual speed, with the isolation, validation, and review discipline that makes the speed safe to trust.
- Making an existing codebase safe to change — tests, CI, and a working agreement for AI coding tools — then shipping the risky change the team has been deferring.
- Product and venture strategy, including technical diligence and build-versus-buy decisions.
- Market, brand, go-to-market, and financial analysis for early-stage ventures.
- Source-grounded knowledge and intelligence systems where accuracy, provenance, privacy, and human review matter.
- Evaluation and operating design: auditing the benchmarks and answer keys a decision depends on, plus failure modes, costs, logging, governance, and what a team needs to run the system after launch.
How I work
Most projects start with one consequential problem, imperfect context, and too many plausible directions. I narrow the opportunity, connect the business case to a buildable product, and put something testable in front of real users. Quality, latency, cost, privacy, and failure modes become visible early; brand, market, and operating questions stay connected to the technical work rather than becoming a later handoff.
Prototype to release-candidate path
A useful first version is more than a demo that works on friendly inputs. I build the product path and the system together: scope, model behavior, retrieval, tests, deployment, and the handoff plan.
Product, market, and knowledge systems
Useful systems begin with the workflow and user, then connect positioning, source quality, model behavior, and what happens when the system is wrong. I keep those decisions in one loop so product and market problems surface before launch.
Architecture, evals, and what to log
Once a system has more than one model or service, the hard part is rarely the model. It is what calls what, what gets logged, what needs human review, and what becomes an audit problem later.
Working alongside your team
I can build independently or join an existing product or engineering team for the AI-specific work. The useful version is staying close enough to ship, then handing off the prompts, eval set, docs, and cost assumptions cleanly.
How engagements work
Start with the smallest engagement that can answer the important product or venture question.
Product + Architecture Sprint
1–2 weeksI examine the opportunity, users, workflow, data, market, and existing systems, then produce a build and validation plan with explicit go/no-go gates.
Prototype Sprint
2–4 weeksA working version in your environment, narrow enough to prove value. Includes evaluation, cost assumptions, and an honest read on whether to harden or stop.
Build + Ship
4–10+ weeksI stay with the build through integration, testing, and handoff. The goal is a system your team can operate without depending on hidden prompt tricks or heroics.
When the question is whether an evaluation can be trusted, the same shape applies. These come out of the research.
Evaluation Audit
1–2 weeksFor an eval suite or vendor benchmark a decision depends on. I establish whether its key is sound and whether its results are artifacts of the items rather than the models. You get a graded findings register, not a suspicion ranking.
Gold-Standard Design
2–4 weeksA keyed evaluation set for a domain where being wrong is expensive, built against primary authority with per-item provenance and machine-checkable field semantics.
Evidence & Provenance Design
2–4 weeksA defensible chain of custody for AI-assisted review: what is mechanizable, what needs a named expert, and what must be attestable afterwards — for a regulator, an auditor, or a deposition.
About
I'm Gregory O'Connor, an MIT-trained Ph.D., product builder, strategist, and co-founder of Socure, based in New York. I build applications and evaluate ventures at the point where technology, product judgment, and business reality have to agree.
My current portfolio includes BNOW.NET, SceneFiend.app, Dreamlet.Studio (co-founded with Jacqui de Jong), advisory and product work with NaviaAI.com, and a stealth AI-adjacent venture under the working title Epistemic. Across that work, I also handle market, brand, go-to-market, and financial analysis—not only the software build. I also do research on evaluation integrity — auditing the benchmarks and answer keys AI systems are graded against — with two papers currently under peer review.
I'm also a working actor with IMDb credits. That grounding in performance and film is part of why SceneFiend and Dreamlet come from inside the craft — how actors choose and prepare material, and which stories film chooses to center — rather than generic entertainment AI.
For stealth and client work, public descriptions stay at the level needed to establish capability. Private methodology, client information, and internal analysis remain private.
Send the workflow
A first email does not need to be polished. Tell me the opportunity, who it serves, what already exists, what data or systems are involved, and what would make a first engagement worth doing.