Vishnu Sai

Toronto, ON ...

vishnu.sai@mail.utoronto.ca
Resume

Backend Engineer with a long-term focus on distributed systems and applied AI use cases

I build backend system and async pipelines for AI and high-throughput workloads.

Outside code, working out, long cycling rides, and reading web novels, are my reset button.

Current focus:

  • Backend engineering for cloud-native services and APIs
  • Infrastructure automation with Terraform, Kubernetes, and CI/CD
  • AI-powered backend pipelines with reliable production integrations
  • Long-term: distributed systems and infrastructure engineering

Engineering highlights:

  • Reduced AI study-kit preparation from days to minutes by redesigning ingestion and generation flows.
  • Designed async job-queue pipelines that decouple heavy AI processing from request lifecycles.
  • Built multi-tenant backend systems with strict tenant-level data isolation and role-aware access.
  • Automated cloud infrastructure for multiple client environments using Terraform and CI/CD.

Experience

Projects

8 total

BeePrepared

WINNER

AI study system that turns long lectures into exam-focused outputs with async processing.

Jan 2026

Final evolution of EduFlow: lecture videos, slides, and notes in, quizzes/flashcards/notes/mock exams out.

My role: Owned backend architecture and async concurrency pipeline in FastAPI.

Turn a 2-hour lecture into a complete study pack in seconds.

AIEdTechBackendCloud

What I built:

  • Rebuilt the original concept around asynchronous ingestion so long lecture jobs no longer blocked API request lifecycles.
  • Designed concurrent background processing for transcription, concept extraction, and multi-artifact generation.
  • Shipped final-exam generation and a cleaner end-to-end UX that made the product demo-ready.
  • Outcome: best version of the concept and the clearest demonstration of backend systems work.

Lumen

WINNER

Mental-health app combining journaling insights with Unity therapeutic minigames.

Aug 2025
Lumen screenshot

Users journal, the app analyzes state, then routes them into supportive interactive experiences.

My role: Built the backend and MongoDB data layer; Unity minigames were built by teammates.

AIHealthTechFrontendProduct

What I built:

  • Implemented backend services for journaling ingestion, user state persistence, and insight generation workflows.
  • Structured MongoDB models and retrieval paths to support fast mood-history lookups and personalized responses.
  • Integrated AI interpretation outputs with downstream app flows while keeping response handling stable.
  • Project won Best Use of MongoDB.

FlowStudio

Agentic video editor that infers user intent and composes edits on top of a base timeline.

Mar 2026
FlowStudio screenshot

Turns raw recordings into edited outputs by combining multimodal context with interaction traces.

My role: Built intent-graph and edit composition logic for backend orchestration.

AIBackendVideoInfrastructure

What I built:

  • Used TwelveLabs plus interaction streams (cursor movement, clicks, key presses, audio, and video context) to infer editing intent.
  • Modeled edits as additive layers referencing a single base timeline to keep composition deterministic.
  • Prioritized tiered rendering (especially cuts) to reduce unnecessary render time and speed iteration.
  • Handled interpolation and timeline alignment so speed/zoom edits composed cleanly on top of prior edits.

ERly

Healthcare triage platform focused on realistic ER wait-time estimation and routing.

Mar 2026
ERly screenshot

Guides users through triage while estimating nearby ER wait times from multiple noisy sources.

My role: Built data/logic for wait-time estimation and triage decision support.

HealthTechBackendMapsAI

What I built:

  • Combined scraped web signals, public APIs, provincial averages, and heuristics into a unified wait-time model.
  • Normalized conflicting signals into practical estimates for nearby ER routing.
  • Integrated triage severity with wait-time predictions so recommendations balanced urgency and practical access.
  • Core challenge was signal quality and calibration, not just API integration.

Wisp

Semantic file-memory system with search, quarantine, and local file intelligence.

Feb-Mar 2026
Wisp screenshot

Indexes file content and metadata into a vector memory so retrieval works by meaning, not just filename.

My role: Built indexing and semantic retrieval systems; did not build the organizer module.

AISystemsProductivityDesktop

What I built:

  • Chunked file content (about <=600 tokens per segment) and stored embeddings with metadata in a vector database.
  • Implemented semantic retrieval so files could be found by actual content meaning, not only path/name filters.
  • Supported quarantine/delete recommendations by combining semantic signals with metadata freshness.
  • Demo failure lesson: index warm-up was missed for the showcase folder, causing a loading-screen live demo.

Temper

Trade-review engine that scores decision quality chess.com-style across windows of activity.

Feb 2026
Temper screenshot

Analyzes local trade windows and full account context to label decisions from brilliant to mistake.

My role: Built processing and heuristic evaluation pipeline with pandas; not a domain trader.

FinTechBackendAnalyticsAI

What I built:

  • Processed tens of thousands of trades with pandas to compute rolling-window and account-level signals.
  • Applied heuristic scoring with labels like good/best/brilliant/mistake/inaccuracy plus AI feedback summaries.
  • Built timeline visualizations, journaling support, and stats views around the analysis pipeline.
  • Scaling lesson: live judge dataset had about 200k trades while rendering and compute paths were tuned for about 20k.

EduFlow AI

WINNER

First version of the study-kit concept that later became BeePrepared.

Oct 2025
EduFlow AI screenshot

Prototype for turning lectures and notes into study outputs, but with weaker generation quality and reliability.

My role: Built the early backend on Next.js API routes before moving to FastAPI in BeePrepared.

AIEdTechBackendAutomation

What I built:

  • Used synchronous API-route style processing, which struggled on larger lecture workloads.
  • Core concept worked, but generation consistency and latency were rough compared with BeePrepared.
  • Served as the experimentation base that directly informed BeePrepared architecture decisions.
  • Clear takeaway: right idea, wrong backend model for scale.

Job Buddy

WINNER

Job tracking and outreach assistant for faster applications.

Jul 2025

Aggregates role opportunities and helps draft personalized recruiter outreach.

My role: Built backend workflows for ingestion, tracking, and outreach generation.

AIAutomationCareer TechBackend

What I built:

  • Automated new role discovery and consolidated updates in a unified dashboard.
  • Generated tailored outreach drafts using profile and resume context.
  • Improved application tracking and reduced repetitive manual job search steps.

Skills

Languages

TypeScript · Python · SQL · Java · C · Bash · PowerShell · JavaScript

Backend

Node.js · FastAPI · Flask · WebSockets · RESTful APIs · Prisma · Express.js · Celery · Asynchronous Programming

Infrastructure

Azure · Terraform · Docker · Kubernetes · GitLab CI/CD · Azure DevOps · AWS · GitHub Actions · Linux

Data

PostgreSQL · MongoDB · SQLite · LanceDB · Vector Databases · Redis · ETL Pipelines

Certifications

Microsoft Azure Fundamentals

Microsoft

Verify credential
AZ-900