G
YouTube TikTok Games
Skip to content
View snayyar00's full-sized avatar

Block or report snayyar00

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
snayyar00/README.md

Hi, I'm Sidharth

Forward deployed engineer. I work with the people who use the software, and I ship the production code.

Based in Canada · open to relocating · sidharthnayyar.com · LinkedIn

AI agents in production: Ares, legal AI (2025 to 2026)

Case volume went from 128 to roughly 1,200 across 25 law firms while I ran engineering, on calls and on site with the firms, shipping their feedback the same week.

  • Agentic core. Tool orchestration, a drafting pipeline, and a medical-chronology engine that turns record sets of up to 10,000 pages into a timeline a paralegal used to spend days on. Per-case cost caps keep a large file from breaking the price.
  • Citation grounding. Every extracted fact is bound to the page it came from, and the agent abstains instead of citing something it cannot find. A lawyer will not file what they cannot verify.
  • Retrieval. Moved the retrieval layer across three vector stores, with an A/B harness to compare strategies before switching. Legal research over case law and the web, with grounded citations.
  • Voice intake agent. Answers personal-injury firms' inbound calls: a 14-step call flow, rules-based lead scoring in under 500 ms to decide a live transfer, and a suite of 50 simulated callers to test every change in batch.
  • Enterprise plumbing. Connectors into SharePoint, NetDocuments, Box and Dropbox that keep each firm's permissions, and a document pipeline that resumes after failure instead of restarting.

AI that ships into code I don't own: WebAbility.io (2023 to present)

My own company. Accessibility for 1,697 live sites and 758 businesses, running inside React, Vue and AngularJS apps behind other people's CSPs and CDNs.

  • Screen-reader agents. An agent walks a page by its accessibility tree and drives real VoiceOver and NVDA on CI runners, recording what a blind user would hear.
  • Code-level fixes. The model layer returns the fix itself, in HTML, JSX or CSS.
  • Speed. Cut the widget's first-interaction time from 1,393 ms to 174 ms, because it runs on other companies' pages and counts against their Core Web Vitals.
  • Tools for coding agents. @webability/mcp is a hosted MCP server with 15 tools: scan a URL or HTML, get framework-aware fixes, verify they landed.

npm npm npm Smithery

npx @webability/mcp

Source: snayyar00/webability-mcp

Enterprise data: ICAO (Montreal, 2024 to present)

Helped lead a finance and procurement migration to Oracle Fusion Cloud, and built the monthly journal pipeline in Python on Azure Synapse, reconciled to the live general ledger.

Stack

  • AI: agentic loops and tool use · MCP · RAG with citation grounding · voice agents · evals and simulation · cost control
  • Code: Python · TypeScript · SQL · React · Next.js · Node · FastAPI
  • Data and infra: Postgres · vector search · Cloudflare Workers · AWS · Azure · Docker · GitHub Actions

B.Sc. Computer Science, Simon Fraser University · sidharth15nayyar@gmail.com

Pinned Loading

  1. webability-mcp webability-mcp Public

    Free accessibility MCP server (WCAG 2.2, ADA, Section 508, EAA). Hosted at https://mcp.webability.io/mcp, no API key for scans. Three engines, structured fixes, verify_fix.

    TypeScript 1