TrustCon 2027 · Engineering track · Program Committee
TrustCon's Engineering track is where practitioners show how trust and safety gets built. That means detection pipelines, classifiers, review tools and the infrastructure that keeps them running at scale.
If you join the Program Committee as an Engineering reviewer, these are the kinds of proposals you'll read. Below are the 22 engineering sessions from TrustCon 2026, each with a short note and a link to its page on the official agenda. They range from solo builders to the largest platforms, from classic systems work to new LLM tooling, and from 25-minute talks to hands-on workshops.
Justin PaineEngineering track chair, TrustCon 2027
Joseph Seering KAISTKyle Bowes BumbleXiuduan Fang Precognition LabsJohn Strieker AtlassianGirija Narlikar Google
The track's opening panel. Engineering leaders on what defines T&S engineering now that AI makes software cheap to build, which skills matter, and how to grow the next generation.
Prasanth Murali Meta
Training policy classifiers with far fewer human labels by putting small heads on frozen LLM embeddings and learning from unlabeled data with known positive rates.
Marta Hawkins Vercel
How one engineer took a platform from report-only detection to fuzzy fingerprinting that catches redeployed phishing kits, malware and clones, with graduated enforcement and a frank look at where it falls short.
Katerina Kolomeytseva Bumble Inc
A fine-tuned language model flags broadly, then uncertain cases go to a second stage that weighs metadata and behavior. Covers routing, thresholds and catching precision regressions.
Rahul Daga Atlassian
A tiered detection and enforcement design for platforms without big teams or vendor budgets, including when regex, YARA and behavioral rules are enough and when to reach for ML.
Shiva Chaitanya Gajula Mercari
Building a service that routes marketplace items through physical inspection before delivery. State modeling, idempotent warehouse integrations and human-in-the-loop workflows.
Juliet Shen ROOSTSavannah Badalich DiscordCarolyn Fox Telus DigitalBarrington Lloyd-Lovett NotionAlice Hunsberger Musubi
Teams at different stages of adopting open-source T&S tooling on what is commodity infrastructure, what is real differentiation, and where scarce engineering time should go.
Jeanie Lam
A fast model, a judge model and expert review in a loop that labeled more than 225 million documents in two months on plain batch inference, with sampling and simple quality metrics.
Josh Allen Roblox
Product, infrastructure and model-accuracy lessons from rolling out facial age checks and behavior-based age signals across a very large mixed-age user base.
James Long Roblox
A streaming detection layer between real-time checks and batch jobs, built on Flink with a caching proxy, that flags scaled abuse shortly after it starts.
Alexander Booth GitHub
An LLM reviewer that recommends Terms of Service enforcement and checks its own work with a panel of judge models, plus the offline evaluation setup and production failure modes.
Harry Huang Scribd, Inc.
Two engineers bring PhotoDNA in-house in a month to scan new and historical uploads, and add a matcher that rescans old content whenever new hash sources arrive.
Enakshi Kaul Google
A monitoring blueprint that pairs inference signals with correction rates to catch silent drift in LLM moderation, with attention to cost and latency.
Juliet Shen ROOSTGeorge Vlasto ResolverDan Sexton IWFVaishnavi J Vys
Where hash matching stops working against generated and evasive content, and how teams build lawful data pipelines and validate models when the training data is illegal.
Shalaleh RismaniDr. Renee Shelby
Cross-functional groups trace one real policy commitment through to the evaluations meant to verify it, and map the gaps between the two.
Samidh Chakrabarti Zentropi
Compares self-checks, separate verifier models, small trained classifiers and deterministic rules as real-time guardrails for agents, on latency, cost, consistency and calibration.
Aniket Ajagaonkar Airbnb Inc
Architecture patterns and pitfalls for using LLMs and coding agents in enforcement decisions and T&S tooling at marketplace scale.
Rashmi Raghunandan Block
Evaluation and monitoring for an evidence-grounded investigation agent that answers “unknown” when evidence is thin, and how phantom citations were driven to near zero.
Tom Thorley GitHubNick Frolov JetBrainsRyan KellyWinston Howes OpenAI
The new safety surface that coding agents open up across models, data flows and execution environments, and how several companies are approaching it.
Abhishek Joshi Patreon, Inc
Turning enforcement history (severity, recurrence and time decay) into a calibrated creator-level risk signal that lets teams step in earlier.
Juliet Shen ROOSTJuan Mrad ROOST
Hands-on time with ROOST's open-source Osprey rules engine and Coop review tool, running in ready-made Codespaces.
Zachary Hughes ThornZoe StollNicholas Schwarzentraub ThornTom Quisel MusubiJason Roche Webflow
What runs inline and what runs async, when to block and when to flag, and how to fail safely when a safety service goes down.
Reviewers read and score proposals like these and help shape the TrustCon 2027 program. If this is work you know well, I'd love to have you on the committee.
Apply to reviewReviewer application link goes here.