{
  "site": {
    "name": "Simar Tathgir",
    "email": "simar19tathgir@gmail.com",
    "linkedin": "https://www.linkedin.com/in/simar-tathgir/",
    "github": "https://github.com/simartathgir/",
    "public": true,
    "tagline": "Software engineer moving into product management.",
    "phone": "608 207 7259"
  },
  "home": {
    "headline": "Hi, I'm Simar Tathgir.",
    "sub": "I'm a software engineer at Goldman Sachs, moving into product management to own what gets built and why.",
    "ctaPrimary": "View projects",
    "ctaResume": "Resume",
    "metrics": [
      {
        "measure": "TA wait times",
        "trend": "down",
        "value": "75%",
        "source": "Office Hours Queue",
        "to": "/work/office-hours-queue#s5"
      },
      {
        "measure": "Tax document processing time",
        "trend": "down",
        "value": "99%",
        "source": "Goldman Sachs",
        "to": "/#x0"
      },
      {
        "measure": "Advisor platform resource use",
        "trend": "down",
        "value": "72%",
        "source": "Goldman Sachs",
        "to": "/#x1"
      },
      {
        "measure": "Wellness app weekly retention",
        "value": "80%",
        "source": "Pocket Therapist beta",
        "to": "/work/pocket-therapist#s3"
      }
    ],
    "experienceLabel": "Experience",
    "projectsLabel": "Projects",
    "educationLabel": "Education",
    "contactHead": "Get in touch",
    "contactBody": "Open to associate product manager roles.",
    "description": "Simar Tathgir. Software engineer moving into product management. Experience, projects and education."
  },
  "experience": [
    {
      "org": "Goldman Sachs",
      "role": "Software Engineer",
      "when": "July 2026 to now",
      "note": "Leads delivery of a cloud based 1099 tax ingestion pipeline; removing a legacy API dependency took a critical step from about 12 seconds to about 70 milliseconds. Led product and technical planning to onboard Deloitte as the platform's second automated document sharing vendor, from user requirements and the end to end workflow to prioritising backend, UI, configuration and security work. Built an AI code review product from a command line MVP into a full web experience, with automated reviews, code walkthroughs and GitLab style diffs."
    },
    {
      "org": "Goldman Sachs",
      "role": "Software Engineering Intern",
      "when": "June to August 2025",
      "note": "Scaled Project Disha, a Salesforce service, to support advisor teams three times larger. Resource use fell 72 percent, stress testing took it to more than 4,000 clients per team, and I defined the scalability success criteria and a repeatable validation process to test it against real usage."
    },
    {
      "org": "Deltek",
      "role": "DevOps Intern",
      "when": "Summer 2024",
      "note": "Automated the setup of a business process platform. Deployments ran 20 percent faster with 75 percent fewer manual errors."
    },
    {
      "org": "Speech Processing and Auditory Neuroscience Lab",
      "role": "Undergraduate Researcher",
      "when": "2023 to 2025",
      "note": "Built a machine learning pipeline that classified sound patterns at 83 percent accuracy, and a pipeline for analysing EEG recordings."
    },
    {
      "org": "UW Madison Division of Information Technology",
      "role": "Student Technical Trainer",
      "when": "October 2023 to May 2026",
      "note": "Taught more than a hundred students across disciplines through hands on workshops built around their knowledge gaps, and grew workshop and one on one participation 40 percent through outreach and partnerships with teaching assistants."
    }
  ],
  "education": {
    "school": "University of Wisconsin Madison",
    "degree": "B.S. Computer Engineering",
    "when": "May 2026",
    "note": "GPA 3.6 and Dean's List. Best Capstone Project Award for Pocket Therapist, chosen from more than 40 teams."
  },
  "projects": [
    {
      "slug": "office-hours-queue",
      "name": "Office Hours Queue",
      "tag": "Student support",
      "oneLine": "Students waited 40 minutes for a teaching assistant. We got it to 10.",
      "role": "I led product and infrastructure execution for a team of four: requirements, sprint planning and the rollout.",
      "when": "March 2026",
      "product": "A queueing platform for office hours, with an LLM reading the question and the code before a human does.",
      "team": "Four of us, ten weeks.",
      "status": "live",
      "stats": [
        {
          "value": "75%",
          "trend": "down",
          "label": "wait time for a teaching assistant"
        },
        {
          "value": "85%",
          "label": "of repeat questions never reached a human"
        },
        {
          "value": "150",
          "label": "students in the pilot semester"
        },
        {
          "value": "10 wks",
          "label": "from idea to launch"
        }
      ],
      "sections": [
        {
          "head": "Problem",
          "body": "Office hours had become a bottleneck. Students queued for about 40 minutes to reach a teaching assistant, and the people running the queue had no way to see what was clogging it.",
          "list": false
        },
        {
          "head": "Users",
          "body": "Three groups with different needs. Students who were stuck and losing an evening. Eight teaching assistants working through a queue that never emptied. Instructors who wanted to know what the class was struggling with.",
          "list": false
        },
        {
          "head": "Discovery",
          "body": "I ran discovery with four teaching assistants and the instructors. The pattern was blunt: the same debugging questions arrived again and again, and answering each one from scratch was what made the queue crawl. That finding reordered the roadmap and put triage at the top.",
          "list": false
        },
        {
          "head": "Decisions",
          "body": "Build triage first. Discovery pointed at repeat questions, so deflecting them was worth more than anything else we could ship.\nScope to the semester. The deadline could not move, so I cut the first release to what four people could take from idea to launch in ten weeks.\nOwn it end to end. I wrote the requirements, ran sprint planning and handled the rollout into the pilot course.",
          "list": false
        },
        {
          "head": "Shipped",
          "body": "Submission analysis that reads the student's code from GitLab and drafts debugging hints for the teaching assistant before a human looks at it.\nSemantic grouping, built on Hugging Face models, that clusters similar questions instead of queueing each one on its own.\nA search across existing Piazza answers, so a solved question is not solved twice.\nAn instructor dashboard that turns queue data into something useful for the next lecture.",
          "list": true
        },
        {
          "head": "Results",
          "body": "Wait time went from 40 minutes to 10, a 75 percent cut.\n85 percent of repeat questions never reached a teaching assistant.\nThe pilot ran with 150 students and 8 teaching assistants.\nThe first release shipped in 10 weeks, on the semester deadline.",
          "list": true
        }
      ],
      "gapHead": "Lessons",
      "gap": "One or two honest lessons, in your own words. What surprised you about the discovery work or about scoping to a fixed deadline? If you kept owning this, what would you do next?",
      "leadStat": 0,
      "description": "Students waited 40 minutes for a teaching assistant. We got it to 10."
    },
    {
      "slug": "pocket-therapist",
      "name": "Pocket Therapist",
      "tag": "Mental wellness",
      "oneLine": "A journaling app that works out how you are doing on the phone itself, so nothing has to be sent anywhere.",
      "role": "Solo. I decided what it should do and then built it.",
      "when": "January 2026",
      "product": "An Android app that reads emotional state from journaling, voice and phone sensors, using on device models and Kalman filter sensor fusion, then suggests something that fits the moment.",
      "team": "Solo build, Android.",
      "status": "live",
      "stats": [
        {
          "value": "80%",
          "label": "weekly retention across the beta"
        },
        {
          "value": "10",
          "label": "beta users"
        },
        {
          "value": "On device",
          "label": "nothing leaves the phone"
        },
        {
          "value": "Best capstone",
          "label": "chosen from more than 40 teams"
        }
      ],
      "sections": [
        {
          "head": "Overview",
          "body": "A mental wellness app that works out how someone is doing from their journaling, their voice and signals from the phone, then offers something useful for that moment rather than a generic tip.\nUnder the hood it runs speech recognition, text processing and emotion classification on the phone, and fuses continuous sensor readings with a Kalman filter, so it stays fast and works offline.",
          "list": false
        },
        {
          "head": "Decisions",
          "body": "Keep the model on the device. In a category this sensitive, trust is worth more than the last few points of accuracy. I traded some accuracy for a product that works offline and never sends anything anywhere.\nNever leave the user empty handed. I wrote the reliability rules for speech errors, interrupted sensors and partial data, so a technical failure still ends in something the person can use.",
          "list": false
        },
        {
          "head": "User journey",
          "body": "I designed the journey from journaling to recommendation and rebuilt onboarding and permissions after watching where people dropped out. Asking someone for microphone and sensor access in a low moment is a big ask, so the flow has to earn it before it asks. Navigation stays low friction, and an interrupted session picks up where it left off.",
          "list": false
        },
        {
          "head": "Results",
          "body": "80 percent of beta users came back each week.\nThe beta ran with 10 people.\nNothing a user wrote or said left their phone.\nIt won the Best Capstone Project Award, chosen from more than 40 teams.",
          "list": true
        }
      ],
      "gapHead": "Lessons",
      "gap": "Your lessons from building this, and the now, next and later you would plan if you picked it back up.",
      "leadStat": 0,
      "description": "A journaling app that works out how you are doing on the phone itself, so nothing has to be sent anywhere."
    },
    {
      "slug": "brancli-ai",
      "name": "Brancli AI",
      "tag": "Hiring",
      "oneLine": "An AI hiring platform that made candidate screening 65 percent more accurate.",
      "role": "I defined the roadmap and the recruiter and candidate journeys, and architected the LLM pipelines.",
      "when": "August 2026",
      "product": "A hiring platform where candidates take interactive LLM based interviews and live task assessments, scored semantically for recruiters.",
      "team": "",
      "status": "live",
      "stats": [
        {
          "value": "65%",
          "trend": "up",
          "label": "candidate screening accuracy"
        },
        {
          "value": "5+",
          "label": "enterprise HR teams in the pilot"
        }
      ],
      "sections": [
        {
          "head": "Problem",
          "body": "Recruiter pain points pointed at screening that was inconsistent and easy to game. I turned them into the product: interactive LLM based interviews and live task assessments, so every candidate is evaluated the same way and the results are harder to tamper with.",
          "list": false
        },
        {
          "head": "Decisions",
          "body": "Let the pilot set the roadmap. Feedback from more than five enterprise HR teams decided which candidate evaluation workflows came first.\nMap both sides of the journey. I defined the critical user journeys for recruiters and for candidates before building either.",
          "list": false
        },
        {
          "head": "Built",
          "body": "Interactive LLM based interviews.\nLive task assessments.\nSemantic scoring pipelines built to scale, on React, Node.js, Python and vector databases.",
          "list": true
        },
        {
          "head": "Results",
          "body": "Candidate screening accuracy improved 65 percent.\nThe pilot ran with more than five enterprise HR teams.",
          "list": true
        }
      ],
      "gapHead": "Lessons",
      "gap": "One or two honest lessons from Brancli AI in your own words. What did the pilot teams teach you, and what would you build next?",
      "leadStat": 0,
      "description": "An AI hiring platform that made candidate screening 65 percent more accurate."
    },
    {
      "slug": "asylum-app",
      "name": "Asylum App",
      "tag": "Legal access",
      "oneLine": "An AI assistant that helps people fill out the U.S. asylum application in their own language.",
      "role": "Designed and built the product: the guided form, the explanations and the chatbot.",
      "when": "March 2026",
      "product": "A multilingual web app that explains the asylum form question by question and routes people to the right section as they go.",
      "team": "",
      "status": "dev",
      "stats": [
        {
          "value": "40%",
          "trend": "down",
          "label": "form errors in internal testing, at best"
        },
        {
          "value": "30%",
          "trend": "up",
          "label": "form completion, estimated in usability trials"
        },
        {
          "value": "5+",
          "label": "languages supported"
        }
      ],
      "sections": [
        {
          "head": "Problem",
          "body": "The U.S. asylum application is long, legal and written in English. Many of the people filling it out are working in another language, often under pressure, and every unclear question is a chance to get something wrong.",
          "list": false
        },
        {
          "head": "Users",
          "body": "People applying for asylum who are working through the form themselves. They need to understand what each question is really asking, in a language they read comfortably, without a legal background.",
          "list": false
        },
        {
          "head": "Decisions",
          "body": "Explain, do not just translate. The assistant uses OpenAI models to explain each question in context and in the user's language, at the moment they get stuck.\nLet people ask in their own words. An adaptive chatbot reads what the user is trying to do and takes them straight to the right section, instead of making them hunt through the form.\nJudge it by the form. Success means fewer errors and more completed applications, which is what internal testing and usability trials measured.",
          "list": false
        },
        {
          "head": "Built",
          "body": "Context aware explanations for each question, in real time.\nMultilingual support across five or more languages.\nAn adaptive chatbot that interprets intent and routes users to the relevant section.\nA Next.js web app with server side rendering and Tailwind.",
          "list": true
        },
        {
          "head": "Results so far",
          "body": "Up to 40 percent fewer form errors in internal testing.\nForm completion up an estimated 30 percent in usability trials.\nSupport across five or more languages.",
          "list": true
        }
      ],
      "gapHead": "Lessons and next steps",
      "gap": "It is still in development. What is left before launch, and what have testing and the usability trials taught you so far?",
      "leadStat": 0,
      "description": "An AI assistant that helps people fill out the U.S. asylum application in their own language. In development."
    }
  ],
  "ui": {
    "experience": "Experience",
    "projects": "Projects",
    "education": "Education",
    "resume": "Resume",
    "email": "Email",
    "contact": "Contact",
    "allWork": "All projects",
    "previous": "Previous",
    "next": "Next",
    "read": "Case study",
    "inProgress": "In progress",
    "role": "Role",
    "when": "When",
    "product": "Product",
    "team": "Team",
    "area": "Area",
    "outcome": "Outcome",
    "sections": "Contents",
    "notFoundHead": "Page not found",
    "notFoundBody": "That page does not exist.",
    "notFoundCta": "Back to home",
    "now": "Now",
    "about": "About",
    "emailLabel": "Email",
    "phoneLabel": "Phone",
    "linkedinLabel": "LinkedIn",
    "githubLabel": "GitHub",
    "openProfile": "Open profile",
    "status": "Status",
    "inDev": "In development"
  },
  "projectsPage": {
    "title": "Projects",
    "lede": "Case studies: the problem, the decisions and what changed.",
    "description": "Projects by Simar Tathgir: the problem, the decisions and what changed."
  },
  "contact": {
    "title": "Contact",
    "lede": "Reach me directly, or leave your email and I will get back to you.",
    "pingHead": "Ping me",
    "pingBody": "Drop your email so I can follow up with you.",
    "pingLabel": "Your email",
    "pingPlaceholder": "you@company.com",
    "pingCta": "Ping",
    "pingSending": "Sending",
    "pingDone": "Pinged. I will follow up with you soon.",
    "pingDuplicate": "Already got this one. I will follow up soon.",
    "pingError": "That did not go through. Try again, or email me directly.",
    "description": "Contact Simar Tathgir: email, phone, LinkedIn, GitHub, or leave your email for a follow up."
  }
}