I build production products where the hard parts are not just code: product complexity, broken workflows, ugly integrations, and poor design.
In August 2025, I shipped SynthicAI, a voice AI agent for customer support.
It reached 500+ waitlist users through Reddit, X, and LinkedIn
It failed because
- the market was already crowded
- the problem was too broad
- I did not talk to enough buyers before building
- AI infra cost became real before funding did
It changed how I build. I stopped chasing broad markets and started looking for narrow, expensive problems with buyers already paying for broken solutions.
Most of these are closed source. A few are public or partially public.
Demo recovery and pipeline verification for B2B sales teams.
Tracks what happened after a demo was booked: meeting status, owner, CRM value, follow-up, and unresolved pipeline.
Revenue intelligence for booking-driven healthcare groups.
Connects patient inquiries, bookings, visits, payments, staff handoffs, and recovery actions into one operating view.
AI visibility tracking for payroll software companies.
Shows how ChatGPT, Claude, Gemini, and Perplexity mention, rank, compare, or exclude payroll platforms in buyer searches.
Refund and approval risk tooling for commerce teams.
Helps teams review refund exposure, policy breaks, approval gaps, and money leaving the business.
AI voice support agent.
Handles missed calls, customer questions, and support handoffs when slow replies cost the business.
Christmas AI product.
Tools for gift ideas, cards, room styling, and holiday planning.
Frontend
Next.js, React, TypeScript, Tailwind CSS, Zustand, TanStack Query, Gsap , motion.dev
Backend
Bun, Elysia, Next.js Server Actions, Inngest, REST APIs, webhooks
Data
PostgreSQL, Neon, Drizzle ORM, Prisma, Qdrant
AI
LLMs, RAG, prompt systems, agents, vector search, evals
Integrations
Stripe, Paddle, HubSpot, Salesforce, Shopify, Zendesk, Slack, Google Workspace
- AI agent workflows that use tools, retry failures, preserve state, and escalate uncertainty
- Extraction systems for messy CRM, support, operations, and revenue data
- Hybrid deterministic + LLM pipelines where facts stay source-grounded
- Human review gates when automation confidence is not enough
- Failure-driven evals based on production mistakes, not benchmark theater
For work only, please.
Secondary: duggal@trysapphire.today