Skip to content
View jaygautam-creator's full-sized avatar
🎯
Focusing
🎯
Focusing
2 followers · 5 following

Achievements

Achievement: Pull Sharkx2Achievement: YOLO

Achievements

Achievement: Pull Sharkx2Achievement: YOLO

Highlights

  • Pro

Block or report jaygautam-creator

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
jaygautam-creator/README.md
Jay Gautam — I build AI agents that act on evidence, and the harnesses that prove where they fail.

AI & ML undergraduate (B.Tech CS, 2023–27). I build LLM agents and RAG pipelines, computer-vision models, and applied ML — and I care most about the evaluation: baselines, ablations, confidence intervals, and hand-built test sets that show where a model fails, not just where it shines.


◆  Published research

A Hard Verbatim-Evidence Gate as the Precondition for Multi-Agent Claim Verification — co-authored preprint on Zenodo (DOI 10.5281/zenodo.22978023, Sep 2026). Requiring every verdict to quote a verbatim source span raised the hallucination-catch rate from 60.3% → 82.8% on SciFact (paired Δ +22.4 pp, 95% CI [12.1, 33.3]). Built into Aletheia.


◆  On the Kaggle leaderboard

Competition Rank Standing Status
ARC Prize 2026 - ARC-AGI-2 #98 of 3,152 Top 4% Live · ends 02 Nov 2026
RSNA Knee Abnormality Detection #387 of 6,363 Top 7% Live · ends 22 Oct 2026
ARC Prize 2026 - ARC-AGI-3 #286 of 4,615 Top 7% Live · ends 02 Nov 2026
Google - The Gemma 4 Developer Agent Competition #527 of 3,443 Top 16% Live · ends 02 Dec 2026
Kaggriculture #5,831 of 10,148 Top 58% Final

Public-leaderboard ranks, refreshed daily from the Kaggle API · last updated 22 Oct 2026


◆  Flagship AI systems

Aletheia — evidence-grounded LLM verification RampBrain — agentic AI, company brain for AI agents Dobara — applied ML revenue-recovery agent River — memory-aware LLM companion

RampBrain is private — rampbrain.com is live, and I'm happy to give a walkthrough.


◆  More AI & ML work

Project What it is
05 Support Triage Agent LLM first-response agent for @AppleSupport built from 2.8M tweets — intent, grounded reply, and an escalation decision citing policy. The evaluation is the point: 5-fold CV with bootstrap CIs, a blind control, a second annotator, and ablations.
06 Dyla Visual search for jewellery — CLIP embeddings + FAISS over 7,000+ catalogue photos, evaluated per condition on hand-shot hard phone photos. Live →
07 CuraLens Medical image screening — MobileNetV2 transfer-learning classifiers for oral and skin images with Grad-CAM heatmaps and risk scoring; 0.92–0.95 held-out validation AUC.
08 Real-Time Parking Detection Computer-vision occupancy pipeline streamed to a live dashboard. Top 10 of 500+ at the IIT Madras Hackathon 2023.

Also shipped: DealFlow360 (Odoo Hackathon 2026 finalist — top 350 of 20,000+) · Blackout Protocol (offline mesh network) · Offline UPI Wallet · all repositories →


◆  Toolkit

Python, scikit-learn, TensorFlow, OpenCV, Anaconda, FastAPI, PostgreSQL, Docker, Linux, Git, TypeScript, React, Next.js

LLMs & agents RAG · hybrid retrieval (pgvector) · LangGraph · MCP · Claude / Gemini / OpenAI APIs  ·  ML scikit-learn · LightGBM · Pandas  ·  Vision OpenCV · MobileNetV2 / EfficientNet · Grad-CAM · CLIP / FAISS  ·  Evaluation CV & bootstrap CIs · baselines & ablations · calibration


Portfolio  ·  Kaggle  ·  LinkedIn  ·  Blog  ·  X  ·  jaygautam561@gmail.com

Pinned Loading

  1. Aletheia Aletheia Public

    Aletheia — an evidence-grounded, multi-agent verification framework with a rigorous evaluation harness that improves the reliability of LLM-generated answers.

    Python 1

  2. AssetFlow AssetFlow Public

    Enterprise asset & resource management — asset status as a state machine, DB-guaranteed no double-booking. Odoo Hackathon, July 2026.

    TypeScript 1

  3. river river Public

    A continuous, memory-aware AI companion that keeps the thread — without taking control of it. Consent-based, editable memory.

    JavaScript 1

  4. offline-upi-wallet offline-upi-wallet Public

    Offline-first UPI-style wallet supporting NFC, QR & audio-based payments with local ledger and auto-sync

    Kotlin 2

  5. CuraLens CuraLens Public

    AI-assisted oral cancer screening system using deep learning

    PureBasic 1

  6. KrishiMitraAI KrishiMitraAI Public

    Python 1