Sharvil Lade

about me.

I enjoy building backend systems, LLM applications, and agentic workflows that solve real-world problems.

experience.

• AI Engineer Intern Bangalore, Karnataka (on-site)

  at, kukufm.com

  May 2026 - Present

  • Built an in-house alternative to Weavy and Magnific: a node-based AI canvas chaining image, video, and audio models; the design team now ships 500+ thumbnails and promo assets daily, cutting manual effort by 70%.
  • Owned a Meta Ads Copilot end-to-end on FastAPI and FastMCP, exposing 50+ ad accounts as Model Context Protocol (MCP) tools for creative insights and plain-language budget reallocation.
  • Pitched and prototyped TRIBE v2, a video analytics model scoring ad creatives for predicted engagement before ad spend; back-tested its rankings against live campaign performance, then demoed the validated results to stakeholders.
  • Shipped the editor hand-off stage of an in-house micro-drama video pipeline, packaging generated video, audio stem separation, and background music with an XML timeline config so editors open a pre-built timeline, cutting edit time by 30%.
  • Instrumented role-based agents with Langfuse, tracing tool-selection accuracy, per-query latency, and hallucination rate, and converting those traces into targeted prompt fixes for mis-routed tool calls.

  • FastAPI
  • FastMCP
  • MCP
  • Langfuse
  • Python
  • AI/ML

• Technology Intern Bangalore, Karnataka (on-site)

  at, reckonsys.com

  February 2026 - April 2026

  • Engineered an event-driven FastAPI microservice for an agentic Q&A product, brokering user queries through Apache Kafka so specialist AI agents resolved them asynchronously without blocking request handling.
  • Orchestrated stateful multi-agent workflows in LangGraph over one shared Redis memory, eliminating context hand-off between agents across Retrieval-Augmented Generation (RAG) pipelines, pgvector semantic search, and MCP tool integrations.

  • FastAPI
  • Apache Kafka
  • LangGraph
  • RAG
  • pgvector
  • MCP
  • Redis
  • Python

• Technical Intern New Delhi, Delhi (remote)

  at, unstop.com

  October 2024 - January 2025

  • Automated high-coverage test case generation for Unstop's daily coding challenge in Python, producing 1,000+ cases and reducing manual authoring effort by 30%.
  • Automated validation of problem statements, I/O formats, and constraints via LLM-driven checks, reclaiming 6-8 hours of manual review weekly and catching malformed problems before publish.

  • Python
  • LLMs
  • Data Structures
  • Algorithms

projects.

• AlphaHive

  github live preview

  • Built a multi-agent equity research system in LangGraph: a supervisor fans out six specialists in parallel across fundamentals, technicals, sentiment, risk, portfolio, and bear-case analysis, then a lead agent merges their ratings into one verdict.
  • Streamed live agent reasoning over Redis-backed SSE with pgvector filing search and Groww portfolio import.

  • Next.js
  • FastAPI
  • LangGraph
  • PostgreSQL
  • pgvector
  • Redis

• True Feedback

  github live preview

  • Built a full-stack anonymous feedback product on Next.js, MongoDB, and NextAuth.js, collecting responses through shareable profile links guarded by email OTP auth and Zod schema validation.
  • Integrated Google Gemini to flag harmful submissions before delivery and route them to a moderation dashboard.

  • Next.js
  • MongoDB
  • NextAuth.js
  • Google Gemini API
  • Zod

technical skills.

• Languages & Databases:

  Python, C++, JavaScript, SQL, PostgreSQL, MongoDB, Redis, pgvector, Qdrant

• AI & Agentic Systems:

  LLMs, RAG, Embeddings, Vector Databases, Agentic Workflows, MCP, LLM Evaluation, LangChain, LangGraph

• Backend & Frontend:

  React.js, Next.js, Node.js, Express.js, FastAPI

• Systems & Infra:

  Docker, AWS, Git, GitHub, GitHub Actions, Postman

• CS Fundamentals:

  Data Structures, Algorithms, OOP, Database Management System, Operating System, Computer Networks