A
Ayush Sharma
About
Detail
Solo AI builder shipping production LLM systems
India
Contact Ayush regarding:
work
Full-time jobs
Résumé
Jobs
verified_user
0% verified
- ACo-FounderAI Automation AgencyJan 2026 - Current (6 months)• Co-built an automation agency shipping LLM-powered workflows for international clients — chatbots, RAG-based assistants, multi-step agents, and custom integrations. Own end-to-end engineering: scoping, architecture, deployment, reliability.
Claude Code Campus AmbassadorAnthropicJan 2026 - Current (6 months)• Lead the Claude Builder Club at IIT Madras (200+ members); run technical workshops on the Claude API, MCP, tool use, and agent engineering. • Shipped the chapter website at claude-builder-club.vercel.app [GitHub]; manage chapter onboarding, vetted Claude Pro distribution, and developer-relations workflow for student builders.- OFounder & CEOOpportuna.worksAug 2025 - Current (11 months)• Built and shipped a marketing-opportunities platform; grew to 500+ student users as solo founder owning product, infra, growth, and customer outcomes.
- Campus PartnerPerplexityAug 2025 - Nov 2025 (4 months)• Drove Perplexity adoption at IIT Madras; ran sessions on AI search and prompt engineering; surfaced product feedback to Perplexity team.
- PSponsorship CoordinatorParadox, IIT MadrasSep 2024 - Jan 2025 (5 months)• Led a 48-member outbound team for IITM’s annual fest; closed ODOO as title sponsor at Rs. 11,00,000 (~$130K) — full enterprise sales cycle from prospect to contract.
Education
verified_user
0% verified
- GGoogle Prompting EssentialsJul 2025
- Developing Explainable AIDuke UniversityJul 2025
- BS, AI and Data ScienceIndian Institute of Technology MadrasSep 2024 - Current (1 year 10 months)Member: Amartya (Finance Society) • Entrepreneur Society
Projects (professional or personal)
verified_user
0% verified
- RReal-Time Voice Agent on Vertex AIFeb 2026• Designed and shipped a production voice agent on Vertex AI using Gemini Flash 2.5 with dual tool-calling for RAG retrieval and live web search. • Built a three-condition activation state machine (active / silent / dismissed) to manage async session lifecycle and eliminate false interruptions. • Engineered tiered response-length logic, explicit context-window management, and automatic English ↔ Hindi language mirroring to keep voice latency under conversational thresholds.
- DPlumb — AI Talent Scouting & Engagement AgentDeccan AI Catalyst hackathonJan 2026• Built an end-to-end AI recruiting agent: parses JDs into structured requirements, scouts a 120-profile candidate corpus, explains each match with requirement-level citations, and ranks candidates into Recommended / Stretch / Nurture / Pass cohorts. • Designed a 5-signal Interest Score rubric (specificity, forward commitment, objection handling, availability, motivation) — each scored 0–20 with cited transcript spans. A production-grade evaluation framework directly tied to recruiter outcomes. • Implemented 4-turn simulated outreach with hidden persona state and live recruiter follow-ups that re-score the expanded transcript — full async re-evaluation loop with state management. • Shipped to production on Vercel + Supabase; supports JSON/C
- EEmpirica — Self-Correcting AI Research EngineDec 2025• Closed-loop hypothesis → experiment → learning system using Gemini thought signatures, generalizing across optimization, materials science, and chemistry domains. • Designed multi-model routing: Gemini 3 Pro for deep reasoning, Flash for fast checks — balancing cost, latency, and output quality per step. SSE streaming pipeline to a real-time reasoning visualizer.