Skip to content
View Sasisundar2211's full-sized avatar
🎯
Vouqis Verify, A AI change verification Platform
🎯
Vouqis Verify, A AI change verification Platform

Highlights

  • Pro

Block or report Sasisundar2211

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. Markdown is supported. 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
SasiSundar2211/README.md

Sasi Sundar

I build developer tooling for AI systems and write about what breaks.

Final-year B.Tech in Artificial Intelligence & Machine Learning, graduating 2027. Everything below links to source. If a number here cannot be reproduced by cloning the repo, it does not belong here.


Building

Vouqis Verify — CI tooling for AI-generated code

A Python CLI on PyPI and a composite GitHub Action. Detects changes on configured AI paths using a git diff, runs the evaluation command the repository already has, and posts a structured review report to the pull request.

87 tests across 9 modules · versions 0.1.1 and 0.1.2 on PyPI · MIT · no users yet

pip install vouqis-verify
vouqis init && vouqis verify

ProcureGuard — vendor drift detection service

FastAPI service that detects procurement price drift and explains its findings. Seven handlers, including async detection dispatch with task-status polling. Gemini 1.5 Flash for detection reasoning, the OpenAI API for the explanations a human reads.

7 API handlers · 2-stage Docker build · Pydantic contracts · pytest · MIT

The task store is in-memory. A restart loses in-flight work, and a second replica breaks status polling. Fine for one instance, not a queue.

NYC Energy Star Predictor — regression with interpretability

Five model families compared under 5-fold cross-validation, the winner tuned over a 40-trial randomised search, then interpreted with LIME and a depth-3 surrogate tree.

9,535 records · 43 features · holdout MAE 8.6064 · R² 0.824

The interpretability layer found a duplicate one-hot column caused by a trailing space in the borough field. Cross-validation never noticed. That bug is the reason I now treat interpretability as a debugging tool rather than a reporting requirement.


Currently

  • Making the Vouqis Verify evidence report useful rather than merely present
  • Re-running the energy pipeline with the whitespace fault fixed and the baseline stored
  • Learning evaluation design for non-deterministic systems, and MLOps fundamentals

Stack

Python · FastAPI · Pydantic · pytest · scikit-learn · pandas · Typer · LIME TypeScript · Next.js · Tailwind · Supabase Infrastructure · Docker · GitHub Actions · PyPI · Vercel Models · Gemini 1.5 Flash · OpenAI API

Learning, not yet in a repo I would point at: PyTorch, SQL, distributed systems.


Writing

I publish roughly one piece a fortnight, usually about a bug I did not expect or a decision with a real trade-off.

  • The shallow clone that made my CI tool pass when it could not run
  • The interpretability layer that found the bug my metrics hid
  • The GitHub App I claimed to have shipped, and had not

sasisundar.me · RSS


Elsewhere

LinkedIn · sasisundar.me · sasisundhar2211@gmail.com

Open to AI Engineer, Applied AI Engineer, ML Engineer and Developer Tools Engineer roles for 2027. Based in India, UTC+5:30.


Every claim here links to the code.

Pinned Loading

  1. Autonomous-Procurement-AI-System Autonomous-Procurement-AI-System Public

    FastAPI service for procurement price-drift detection and vendor ranking. Gemini drift summaries, OpenAI ranking explanations, Docker, React dashboard.

    TypeScript

  2. VouqisVerify VouqisVerify Public

    Vouqis Verify (VouqisVerify) generates AI Change Evidence Packs from merged GitHub pull requests, with AI change classifications, CI checks, and reviewer evidence exported to Excel and CSV.

    TypeScript

  3. Cognis Cognis Public

  4. ml-project_implementation ml-project_implementation Public

    End-to-end supervised regression on 9,535 NYC buildings. IQR outlier removal, log-transform, collinearity pruning → 5-model CV → two-stage hyperparameter search → GradientBoosting MAE 8.6 R²=0.82. …

    Python