i build things in domains most teams won't touch. financial compliance, regulatory interpretation, wealth management, healthcare pre-authorization. domains where hallucination isn't a demo bug, it's a regulatory violation.
the most interesting thing about AI isn't that it can think. it's that it makes you realize how little of your job was thinking.
so i build the parts that actually require thinking.
|
fully autonomous content engine for X. claude opus writes, scores, schedules, publishes, and self-replies. zero manual intervention. not a tweet scheduler with AI sprinkled on. a complete autonomous content architecture with voice calibration, territory balancing, algo-aware timing, and engagement loops. the question isn't whether AI can write tweets. the question is whether it can develop taste. |
benchmark-optimized agentic coding CLI. forked from codex, closing 9 specific gaps vs claude code. plan-act-verify loops. reflexion-on-failure. worktree isolation. three-layer evaluation system targeting SWE-bench Verified and Terminal-Bench 2.0. spent 36 hours automating a process i had never once done successfully by hand. if you need someone to scale nothing to production, i'm available. |
two open-source CLAUDE.md systems that turn claude code into a full company. not code generation. strategy, design, engineering, compliance, growth, monetization.
|
Claude Web Studio -- 28 specialist agents
the orchestrator is prohibited from writing code or making decisions. it can only route to agents. this solves the #1 problem with multi-agent systems -- the orchestrator doing everything itself. |
Claude Mobile Studio -- 26 specialist agents iOS (SwiftUI) . Android (Jetpack Compose) . React Native . Flutter idea to app store. ASO, push notifications, monetization, compliance (ATT, Privacy Manifests). same eval system. same negative permissions architecture. |
| what it does | |
|---|---|
| Boredfolio | india's most transparent mutual fund platform. ai-powered fund roasts, ELSS comparisons, direct vs regular calculators. every joke contains a real financial truth. next.js + claude. |
| Jugalbandi | dual-interface wealth platform. client dashboard and RM command center on a shared intelligence graph. the RM doesn't need instructions -- they need context. |
| Sarvam BFSI | 54 AI use cases for indian banking across 8 languages and 5 sarvam APIs. compliance surfaces that earn trust. |
| RegAI | RAG pipeline for SEBI and RBI regulations. domain-specific chunking that preserves table context and amendment chains. claude + pgvector. |
| PreAuth AI | 5-node AI engine for hospital insurance pre-authorization. clinical parser, ICD-10 coding, completeness checker, approval scorer. days to minutes. |
| Resume Scorer | 6-dimension AI resume scoring. semantic skill matching, bias detection, explainable reasoning. |
| Job Autopilot v3 | 7-stage AI outreach pipeline. scrape, guardrails, score, compose, resume gen, send, monitor. 2,500+ lines of python. zero manual steps from JD to personalized application. |
| The Delusional Manifesto | 22,000 words on AI-native personal finance in india. 10 chapters. 5 product specs. full system architecture. |
| Agentic WealthOps | 52 use cases for AI agents in indian broking, MF distribution, and insurance. |
| Claude for Indian BFSI | claude go-to-market playbook for indian banking. RBI compliance mapping, enterprise use cases, production prompt templates. |
| LLM Eval BFSI | benchmark framework for LLMs on indian financial services tasks. claude vs GPT-4 vs gemini on regulatory QA. |
- anthropics/financial-services-plugins#15 -- indian BFSI plugin
- anthropics/claude-cookbooks#407 -- domain-specific RAG cookbook
| claude-on-a-budget | token compression for claude code. real counting, semantic-safe transforms, behavioral regression testing. |
| unga-bunga | every message you send to claude code has a hidden tax. this shows you the bill, then fixes it. |