I build AI systems that connect models to evidence, people, and real tasks.
Applied AI · Retrieval · Agent Reliability · Workflow Systems · Evaluations · Agent Reliability ·
📍 New Delhi, India · Currently Applied AI Engineer at Bynd AI and freelancing
- 🧭 canon — Policy-governed AI that retrieves approved rules, applies constraints, and returns evidence-backed answers.
- 🗂️ claris — Provenance-aware clinical case investigation for HCC research.
- 🧾 Tender Compliance Engine — Extracts tender requirements with source links for review and export.
- 🧭 atlas — Synthetic-data dashboard prototype for exploring operational metrics.
- 🧰 lintel — Local-first linter that screens AI agent repositories for reliability signals and produces an evidence-linked report.
- 🛡️ agentic policy reliability — Local demo of dated policy lookup, clause evidence, abstention, and human-review routing.
- 🎯 tendermatch — Finds and ranks procurement opportunities for specialized suppliers.
- ♻️ medwaste guardian — Multimodal prototype for biomedical waste classification and compliance guidance.
- 🛡️ clara agent reliability — Demo pipeline for grounded claims, risk checks, review gates, and trace logging.
- 🌱 AURA 2.0 — Exploring consent-based, multimodal assistance with clear limits around mood and emotion inference.
- 🗣️ aura — Voice-first, vision-assisted help for understanding objects and written information.
- 💛 Ally — An experiment combining speech, language, and visual cues for emotionally aware interaction.
- 🩺 SwasthaSamvad — A source-grounded medical information prototype with human-review and safety boundaries.
- 🤖 Baymax — A fine-tuned language-model project with reproducibility and evaluation requirements documented in the repository.
- 🦚 ask krishna — A conversational guide to the Bhagavad Gita, built around retrieval and generated responses.
Experiments and works in progress with room to grow into stronger public demos.
- 🧠 hydran — Reproducible brain MRI segmentation research with a PyTorch training setup and reported evaluation results.
- 🎯 Tender Intelligence Engine — Procurement intelligence workflows for discovering and assessing tenders.
- 🔬 OEM Intelligence — Evidence-backed research and matching for scientific instrument opportunities.
- 📄 OCR Benchmark — Experiments comparing OCR and document extraction workflows.
- 🔄 Parity — Early work on reconciling messy business data into trusted decisions.
- 💬 Socratic Bot — A guided-learning chatbot prototype built around conversational problem solving.
- 🎙️ siri + n8n assistant — Connects Siri Shortcuts to email, calendar, and research workflows.
- 🖱️ gesture-controlled virtual mouse — A computer-vision interface controlled by hand gestures.
- 📚 machine learning practice — A collection of machine learning experiments and implementations.
- 📈 tradesync — Compares rule-based trading signals with an XGBoost approach.
- 🧰 issue-management — Read-only workflows for Linear review and standup context.
Selected merged contributions to open-source projects:
- 🛰️ agentglass — Dashboard for coding-agent activity, tokens, and tool calls. Improved CI install behavior in a PR.
- 🌐 webclaw — Web extraction for LLMs and agents. Clarified PDF empty-content behavior in a docs PR.
- 🔍 agent-inspect — Local evidence and trajectory debugging for TypeScript agents. Added an anonymized case-study template in a docs PR.
- ✍️ commit-echo — Learns a developer's commit style and suggests messages. Fixed config updates overwriting the template path in a PR.
- 🧩 triagekit — Generates GitHub issue templates. Fixed generated issue configuration in a PR.
- 🧭 retinue — A compose-based personal-agent platform. Corrected image update steps in a docs PR.
I write about:
- retrieval systems
- evaluation pipelines
- grounding failures
- orchestration tradeoffs
- operational AI reliability
