Overview
The Problem: $743B in e-commerce returns annually and most low-value damage claims still take 3 days of human back-and-forth. Existing AI agents (Sierra, Decagon, etc) automate the customer dialogue. ClaimsForge is the Trust Layer that compliance and merchant ops teams have been waiting for.
How It Works: Seven Gemini 2.5 specialist agents run as a parallel asyncio pipeline: IntentAgent → (Emotion ‖ Needs ‖ DamageAgent) → CompensationAgent → SupervisorAgent → VerifierAgent.
Customer uploads a damaged item; DamageAgent (Gemini 2.5 Vision) returns bounding-box-localized severity + detected_subject. CompensationAgent picks from 26 policies + 1,329-entry hybrid-RAG KB (Gemini embedding + keyword). Then an AWS-IAM-style Supervisor enforces deterministic hard rules in pure Python: pHash visual-replay gate, EXIF age check, multimodal text/image consistency, duplicate-order detection, $500 cash cap, perishable exemption. Every reply ships with a Stripe-Radar-style Trust Score, 6 weighted factors (image uniqueness, image provenance, amount sandbox, history coherence, emotion gating, evidence quality), each with a rule_id backlink to the supervisor decision that drove it.
What Makes It Different • 100% Auditable Decisions — every rule is pure Python, every match logged with rule_id (no opacity) • Deepfake-Aware — pHash collision detection + EXIF DateTimeOriginal sanity check + multimodal text/image consistency gate (99% of insurers have seen AI-tampered evidence; only 32% feel confident catching it) • Quick Self-Deploy — MIT-licensed, single Vultr VM, no managed-service contract (Sierra typical deploy is $200K-2M ARR) • Self-Evolving — 47 methodologies auto-distilled by Gemini from resolved cases • Tier-2 Hot-Editable Rules — ops team toggles "Liquid on Electronics escalates" via admin UI, no code deploy • Data Flywheel — every approved claim auto-labeled gold/normal/red_flag, one-click SFT export to Vertex AI / OpenAI / Anthropic fine-tuning formats
Note: the demo URL is substituted with the GitHub repository link.
Tech stack
No comments yet
Be the first to comment on this project.
sentinel
Full-Stack Self-Healing DevOps & Security Copilot The Concept: An AI operations engineer that doesn't just suggest code fixes—it monitors cloud infra, detects uptime/security drops, automatically provisions fixes, and generates audit reports for enterprise compliance.
Agency Compiler
Agentic creative project compiler. Most creatives, agencies and designers dont know how to use ai, none the less agents. We created a platform that allows a creative agency or studio owner to instantly spin up agents based on the project ask, resources and scale.
ProofRun
ProofRun helps developers answer a critical question before shipping: will this update break behavior users rely on? An AI agent inspects two Git revisions, designs targeted experiments, and proposes repairs. An independent Docker-based verifier runs those experiments against both versions, preserves the original tests, and checks whether the repair actually works. Developers receive an update, skip, or postpone recommendation with recorded evidence and a reviewable patch. In a live synthetic…