Sahil Regonda
Full-stack and AI engineer focused on reliable product delivery.
I study Computer Science at the University of Toronto, with minors in Mathematics and Statistics, and graduate in May 2028. Across five engineering internships, I've worked on social fitness, multi-tenant SaaS, founder tools, vehicle verification, and multi-store operations.
My work spans product interfaces, backend services, data contracts, permissions, integrations, testing, and ML deployment. I am most effective where those layers have to behave as one dependable system.
That has meant recovering expired sessions without losing useful state, enforcing privacy rules from the API through PostgreSQL, preserving evidence in financial research, and making trained models smaller without changing their output.
Open to Summer 2027 software engineering internship opportunities.
Notebook
Sketchpad. Draw, drop in an image, save what you make.Pen · shortcuts: B M H A C E for brushes, L R U O for shapes, G I T S for fill, pick, text, stamp · [ ] resize · Cmd+Z undo · Cmd+S save
Experience
Where I've worked and what I shipped.
AI Engineering Intern
Seagulls Labs — OpenStage
Remote
Improving reliable entry and workout completion for a privacy-aware social fitness platform.
Hardened four authenticated entry surfaces with expired-session recovery, in-place retries, and partial-failure preservation so useful content remains available when a secondary request fails.
Deferred Gym history and statistics and scoped Settings requests to the screens that use them, removing up to 12 unnecessary initial catalog requests in recorded local traces; cold and warm localhost p95 remained below 1.3 seconds across 200 loads.
Software Engineering Intern
Applied Optimal Inc. — ClubConnect
Remote
Shipping membership, roster, activity, and permission workflows for a multi-tenant club-management platform.
Delivered 13 initiatives through 28 merged pull requests across the Flutter client, FastAPI services, PostgreSQL migrations, and Supabase policies.
Protected birth dates and tenant data through role-aware APIs, row-level security, and server-authoritative age filtering, including leap-day and time-zone handling.
Team/product context: ClubConnect launched to 100+ users during this period.
Software Engineer Intern
Inclusifai
Toronto, Ontario
Brought meetings, calendars, reminders, and reliable date handling into the founder funding workflow.
Embedded Jitsi meetings inside funding-opportunity cards with one-click join, link sharing, lifecycle cleanup, and return-to-opportunity behavior.
Completed Google and Outlook OAuth calendar synchronization with scheduled reminders and integration testing across stored meetings, tasks, and deadlines.
Software Engineer Intern
PashMotors — VINSecure
Remote
Improved the reliability of vehicle registration, verification, camera capture, and VIN workflows.
Debugged registration, camera, and dashboard failures across five browsers and five documented device and operating-system environments, contributing JavaScript, Node.js, and responsive interface fixes.
Tested verification with four email providers and traced provider-specific failures across frontend state, provider behavior, and API responses.
AI Engineering Intern
Ontario Inc.
Windsor, Ontario
Built shared operational workflows for paperwork, inventory, schedules, and delivery orders across 5+ stores.
Built React and TypeScript interfaces with FastAPI and PostgreSQL services for paperwork, inventory, schedules, orders, and multi-location records.
Co-led a DoorDash API integration with another developer, connecting delivery-order data to React and PostgreSQL tablet workflows with validation, error handling, and end-to-end tests.
Projects
Built end to end, from model training through to the deployed interface.
AccessiGo
A shared web and mobile system for classifying entrance images.
AccessiGo is a team project that distinguishes entrance doors from non-entrance doors and serves results to React and React Native clients. It provides entrance information; it does not certify accessibility, safety, legal compliance, or current availability.
I helped move the service from Flask/WSGI to FastAPI/ASGI, moved image preprocessing into memory, reused the inference session, and separated stateless scoring from authenticated persistence. Exporting the TensorFlow model to ONNX reduced the artifact from 44.4 MB to 14.8 MB, with an eight-input synthetic parity gate at 1e-5 tolerance before deployment.
SEC Document Copilot
A filings research assistant that cites its sources or reports that evidence is missing.
SEC Document Copilot is an authenticated research workspace over 25 annual filings from Apple, Microsoft, NVIDIA, Amazon, and Alphabet. It keeps search, source inspection, citations, financial context, and saved research in one place.
Deterministic SEC ingestion preserves financial row labels, reporting periods, units, source locations, and artifact provenance. Retrieval combines pgvector semantic search with PostgreSQL full-text search through Reciprocal Rank Fusion, while registered evidence, citation and value checks, one bounded retry, and an explicit no-evidence fallback keep unsupported answers out of the result.
SkillSnapAI
Resume analysis and ATS-oriented feedback matched to a target role.
SkillSnapAI accepts a PDF resume and job description, then returns structured feedback on alignment, keywords, readability, tone, and ATS compatibility. Authenticated users can review earlier analyses, preview uploaded resumes, and delete their stored files and records.
Pathfinder Mastery
An interactive comparison of BFS, DFS, Dijkstra, and A*.
A pathfinding visualizer with ordered waypoints, segment-by-segment routes, coordinate controls, undo, and restart. The browser version ports the original Java Swing application to React and TypeScript while preserving the algorithms and interaction model.
Trial of the Knight
A 2D platformer in Python and Pygame.
A knight moves through shifting magic terrain, traps, and room puzzles. Written without an engine, so the game loop, collision handling, and sprite state machines are all hand-rolled.
Dungeon Runner
A dungeon crawler running on plain HTML, CSS, and JavaScript.
Procedural mazes and puzzles with no framework and no build step. Rendering, input handling, and level state are written directly against the DOM and canvas.
Contact
Interested in working together? The best way to reach me is by email or LinkedIn.
Open to Summer 2027 software engineering internship opportunities.