Sashank Tadimeti
Prev 2x GenAI SWE Intern @ Fidelity Investments
Georgia Tech BS/MS CS major building AI/ML systems. Also a Database Systems (CS 4400) Teaching Assistant.
I'm an AI/ML engineer building production-ready GenAI + retrieval systems, agentic AI workflows, and scalable data/ML infrastructure. I care about shipping products that create real impact at scale.
Currently, I'm pursuing a BS/MS in Computer Science at Georgia Tech—Bachelor's specialization in Artificial Intelligence and Information Internetworks, Master's in Machine Learning. I'm a Database Systems (CS 4400) Teaching Assistant, and previously completed 2x GenAI Software Engineer internships at Fidelity Investments. I'm also the co-founder of EchoBoard, an AI-driven survey analysis platform that automates feedback processing and generates actionable insights.
My work focuses on AI/ML systems, scalable data infrastructure, and distributed computing. I've developed production-grade GenAI search engines, multi-agent orchestration systems, and ML pipelines—from deploying RAG systems handling 200K+ records to building autonomous agent workflows that reduce research time by 80%. My projects span multi-agent orchestration, code generation and automation, LLM-driven extraction and analysis, ML pipelines (NLP, embeddings, clustering), diffusion models and generative AI, and scalable backend infrastructure.
I'm passionate about building impactful things that solve real problems. Whether it's automating complex processes, designing multi-agent systems, or creating tools that improve workflows, I focus on shipping solutions that make a measurable difference.
Teaching Assistant for CS 4400 (Database Systems) at Georgia Tech, supporting students in relational database design, SQL optimization, transaction management, and distributed systems.
Developed GenAI search engine using Microsoft GraphRAG for 200K+ research records, cutting research time by 5+ hours/week and improving graph generation fidelity by 30%, while optimizing real-time inferencing with LanceDB achieving ~10s latency and building Streamlit UI with REST APIs powering 500+ daily queries.
Founded EchoBoard (echoboard.us), automating AI-driven survey analysis and cutting manual effort by 85% through LLM insights pipeline generating 90%+ data-backed insights, built full-stack application with Flask, React, and Supabase for customizable dashboards, and deployed via Vercel ensuring <2s page loads with integrated AI Assistant for natural-language querying.
Built immersive photogrammetry museum in Unity for Meta Quest, scripting dynamic scavenger hunts with live score tracking and integrating 3D interactions via XR Interaction Toolkit.
Deployed full-stack semantic search bot to production for internal risk site using Python, Angular, AWS, and FastAPI, integrating RAG with Hugging Face LLM API for contextual text generation of 20+ articles and creating article recommendation pipeline with SBERT achieving 96% precision, while refactoring Jenkins CI/CD pipelines with Docker & Kubernetes.
Cloud Vibecoder
Source CodeMobile AI code platform that turns ideas into plans, executes agent loops, and generates GitHub PRs with code changes. Designed multi-agent orchestration using OpenAI/Claude LLMs with plan generation, tool-calling, and repo-aware file edits, achieving 91% valid PRs and <3 min avg idea-to-code latency.
Competitive Intelligence (CI) Assistant
Source CodeAutonomous, multi-agent CI system integrating LLMs, web scraping, and public APIs for real-time market analysis. Reduced research time by 80% by automating intelligence gathering from idea validation to strategic recommendations. Built interactive platform (React, FastAPI) for real-time visualizations and dynamic exploration of competitive landscapes.
GenETF
ReportML data pipeline for optimized ETF portfolio creation using NLP preprocessing and ModernBERT embeddings. Applied PCA to reduce embeddings (1024 to 349 dims) maintaining 95% variance and developed hierarchical clustering to semantically group 5K+ companies. Maximized Sharpe Ratio (2.98) in backtested ETF optimizer using SLSQP with L2 regularization.
Conditional diffusion model with Temporal U-Net/FiLM conditioning to generate coordinated trajectories for 11 players. Applied conditioning with 8 features and classifier-free guidance, achieving 80% LLM approval and 50% coordinator approval. Achieved 48.96% directional accuracy (vs. 36–40% autoregressive baselines) across ADE/FDE, context adherence, and diversity.
Donateables
Source CodeCross-platform mobile app (Java/Android and Swift/iOS) with geolocation features to streamline donation workflows. Designed matching algorithm with category-based semantic understanding, achieving 95%+ accuracy for real-time tracking. Facilitated donation network connecting 200+ users with 5+ organizations, validated by key community stakeholders at app fair.
VisualAid
ReportPortable OCR assistive technology for individuals with macular degeneration, enabling real-time text-to-speech. Utilized OpenCV/Pytesseract/NumPy to create CV system with >97% accuracy in text recognition under 20 ft. Led the end-to-end product lifecycle as Team CEO, achieving 100% satisfaction in a successful pilot program with a key client.