Danish Siddique
Senior Full-Stack Engineer · Agentic AI
I'm a full-stack engineer with 5+ years building products end to end, and lately the AI underneath them. At (formerly Mynd) I built Kindred Practice, a platform used by 200+ therapists, and now lead the memory work in , the AI service every Kindred product calls: memory that can be rebuilt and erased, guardrails that fail closed, and tracing on every call. I also built the agents that turn Sentry alerts and tickets into reviewed pull requests.
After hours I run (soonbuilt.com), a small studio that has shipped (tallyreel.com), an app that makes finished videos on your own computer and posts them to 15 networks, and (joinroundzero.com), live voice AI mock interviews. I use Claude Code and Cursor every day to move faster, but not to do the thinking. I still read every line before it ships.
Experience
- Built and launched TallyReel (tallyreel.com): a Mac and Windows app that turns one line into a finished video entirely on-device (Qwen3.5 on MLX/llama.cpp, Kokoro voice, Whisper captions, numpy-composed music, ffmpeg), plus a web and mobile service that posts it to 15 networks from a pg-boss queue.
- Built and launched Round Zero (joinroundzero.com): live voice AI mock interviews on LiveKit and Deepgram, with scorecard quotes verified against the transcript and a coach that remembers every session.
- Built the Soonbuilt site and its idea-search engine on one Cloudflare Worker: parallel web research, a prior-art pass that drops ideas that already exist, and code-level guards that cut unsourced links and figures.
- Build Kindred Workplace, an emotional-intelligence platform used by 8,300+ people across 21 companies: built its conversation-analysis feature from scratch, made the AI coach context-aware with release-gating evals, brought coaching into Slack and Teams, and took it to production on AWS.
- Top contributor to Kindred Practice (formerly Mynd) over three years (4,400+ commits), the practice-management platform 200+ therapists run on: video sessions, AI-drafted session notes, an AI assistant over past sessions, scheduling and billing.
- Cut AI response latency ~40% and model spend ~30% across production RAG pipelines through caching, model routing and eval-driven iteration.
- Lead memory in Kindred Core, the shared AI platform behind Kindred's coaches: memory that updates instead of contradicting itself and deletes a person on request, with scrubbed private data and every AI call traced.
- Built Agent Ops: AI agents that turn crash alerts and tickets into plans a human approves, then pull requests, fixing 30+ issues a month across four product teams.
- Built a RAG-powered customer support chatbot on OpenAI, FastAPI, and PostgreSQL. Improved response accuracy and significantly reduced manual support load.
- Designed and shipped cross-platform React interfaces for the SaaS product.
- Integrated third-party APIs into SaaS workflows on AWS Lambda using a serverless architecture, enabling automation and data sync across platforms.
- Built an ML-based trademark infringement detection system on logos. Reached 85%+ accuracy. Reduced manual review effort by ~70%.
- Delivered multiple applications across Node.js, React, Flask, and PostgreSQL for legal and marketing automation.
- Refactored PostgreSQL and MongoDB databases. Improved query performance and scalability under growing data loads.