AAYUSH
SHARMA
Currently
Experience
GenAI Intern
Delivered cloud-native Generative AI solutions for Fortune 500 clients across design, implementation, deployment, and production reliability.
- •Engineered an end-to-end AI Interview Avatar platform integrating real-time speech processing, LLM inference, APIs, and conversational workflows — processed 250+ technical assessments.
- •Achieved 94% question-relevance accuracy and sub-500 ms end-to-end speech-to-LLM-to-speech latency while reducing operational costs by 35%.
Tech Lead — Deloitte Capstone
Built FairLend, a GenAI-powered fair-lending validation workbench for banks and NBFCs.
- •Built FairLend with FastAPI, Next.js, PostgreSQL, and Google Gemini — cutting audit time from 4–6 weeks to 2–3 days per model run across 3,100+ synthetic profiles and 5 fairness dimensions.
- •Partnered with a 4-member team and Deloitte mentors to align fairness KPIs with RBI-style governance standards, producing audit-grade evidence packs for banks and NBFCs.
Software Engineer Intern
Built production voice-AI and backend systems in a high-scale telecom environment.
- •Built a Voicebot Onboarding platform that reduced enterprise-customer integration time by 27%.
- •Developed and scaled backend microservices with Spring Boot, Next.js, and MongoDB, supporting 10,000+ concurrent requests at sub-200 ms latency.
- •Fine-tuned Speech-to-Text models for Indian languages, improving production transcription accuracy and enabling multilingual voice-AI workflows.
Achievements
Projects
AI Cloud Resource Optimizer
Real-time ML-driven cloud resource prediction & auto-scaling. Reduces infra costs by 20%.
FairLend (Fair Lending AI Workbench)
Deloitte capstone. GenAI workbench that stress-tests education loan models across 5 fairness dimensions and 3,100+ profiles, cutting a 4–6 week bias audit down to 2–3 days.

Tallystick
Collapses six months of model-risk and fair-lending review into a byproduct of scoring. 28-page validation pack in 30 seconds, 226 challengers searched for a fairer model.

Prism (Cursor for PMs)
Cursor's interaction model applied to product decisions. 9 agent tools can rewrite your PRDs, 0 changes land without approval — every edit is a diff, cited to its source.

Voicebot Onboarding Platform
Voicebot onboarding platform that reduced customer integration time by 27%. Built scalable workflows with STT optimization and automated API-driven setup.

ThoughtSpace
Spatial canvas app for rediscovering saved notes across Health, Wealth, and Wisdom. Uses GPT-4 to auto-categorize content and surface meaningful connections.


