Rishika Singh

Rishika SinghAug 7, 2025

Excited to share that I've been selected for the Amazon ML Summer School 2025! Out of 60,000+ applicants, only about 3,000 were selected and I'm incredibly proud to be one of them! The program offers an incredible opportunity to learn directly from experienced ML scientists at #AmazonIndia, covering topics such as: Supervised & Unsupervised Learning Deep Neural Networks Dimensionality Reduction Generative Al and LLMs Sequential Learning Causal Inferencing Reinforcement Learning Looking forward to strengthening my ML foundations and gaining industry-relevant insights through this immersive experience. #AmazonMLSummer School #AmazonIndia #MachineLearning #DataScience #AIResearch #artificialIntelligence #GenerativeAl

Rishika SinghApr 14, 2026

I recently built a fully autonomous AI voice agent demo for a YCombinator startup that handles inbound calls, manages Google Calendar, and logs data in real-time. Moving from "Hello World" to a production-ready voice agent required navigating some significant technical hurdles that weren't in the docs. 🛠 The Technical Deep Dive The architecture relies on a "𝗙𝘂𝗻𝗰𝘁𝗶𝗼𝗻 𝗖𝗮𝗹𝗹𝗶𝗻𝗴" loop. When the agent decides to book a meeting, it triggers a backend hook. Here is how I navigated the primary integration challenges: Retell AI 𝗡𝗲𝘀𝘁𝗲𝗱 𝗣𝗮𝘆𝗹𝗼𝗮𝗱 𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲: A major friction point was discovering that Retell wraps all function arguments inside a nested object. Standard schemas will throw 400 errors until you account for this specific wrapping—a detail I only identified through persistent terminal logging during live calls. 𝗧𝗵𝗲 𝗣𝗿𝗶𝘀𝗺𝗮 𝘃𝟳 𝗠𝗶𝗴𝗿𝗮𝘁𝗶𝗼𝗻: Moving to Prisma v7 was a significant shift. The removal of the database URL from the schema file and the transition toward driver adapters meant my existing v5/v6 workflows were obsolete. Success here required a line-by-line audit of the migration guide to ensure proper type safety and connection management. 𝗧𝘂𝗿𝗯𝗼𝗽𝗮𝗰𝗸 & 𝗦𝗤𝗟𝗶𝘁𝗲 𝗣𝗲𝗿𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝗲: In Next.js 16, native modules like better-sqlite3 behave differently under the Turbopack runtime. I solved this by moving away from environment variable parsing for the database path and implementing absolute pathing to ensure the runtime could consistently locate the local data store. 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗶𝗹𝗶𝘁𝘆: Using ngrok wasn't just for external exposure; it was the backbone of my debugging process. Monitoring the live JSON exchange between the AI's "brain" and my API allowed me to resolve the "technical difficulty" errors that often occur mid-conversation due to schema mismatches. 🚀 The End Result: "Aria" 𝗟𝗼𝗴𝗶𝗰: A voice agent that checks real-time Google Calendar availability and books 30-minute slots. 𝗢𝘂𝘁𝗯𝗼𝘂𝗻𝗱 𝗖𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝘆: A dashboard-triggered system where Aria can initiate calls to any number. 𝗙𝘂𝗹𝗹-𝗦𝘁𝗮𝗰𝗸 𝗩𝗶𝘀𝗶𝗯𝗶𝗹𝗶𝘁𝘆: An admin dashboard featuring auto-refreshing tables for every message and booking stored in SQLite. 𝗧𝗵𝗲 𝗦𝘁𝗮𝗰𝗸: Next.js 16, Retell AI, Prisma v7, SQLite, Google Calendar API, Nodemailer, and ngrok. The hardest part of Voice AI isn't the conversation—it's the reliability.If you're currently wrestling with Retell AI payloads or Next.js 16 edge cases, I'd be happy to connect and share what I learned. I am also open to any suggestions or feedback to make this project even better! Project demo at: https://lnkd.in/gbqqhXB7 #AI #VoiceAI #RetellAI #NextJS #BuildInPublic #SoftwareEngineering #FullStack #AIReceptionist

Rishika SinghMay 10, 2024

Thrilled to announce my selection as a Contributor for GSSOC'24! Joining this empowering program has been an honor, as I look forward to contributing to meaningful open source projects. The journey ahead is filled with excitement, knowing that I will have the opportunity to learn from inspiring mentors and peers. This experience will not only broaden my skills but also allow me to make a positive impact through coding. Let's code for change! #GSSoC24 #OpenSource #CodingForChange