Skip to content
View jeevesh2515's full-sized avatar
💭
Working and learning
💭
Working and learning
6 followers · 11 following

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLOAchievement: Pair Extraordinaire

Achievements

Achievement: QuickdrawAchievement: Pull Sharkx2Achievement: YOLOAchievement: Pair Extraordinaire

Block or report jeevesh2515

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
jeevesh2515/README.md

Jeevesh Singale

AI & Systems Engineer · Voice AI · Agentic RAG · 3D WebGL

I build AI systems you can audit: grounded answers, bounded tools, published evals.

Nottingham, UK  •  LinkedIn  •  Portfolio  •  Email  •  Sponsor


Followers: 6 Stars: 41 Profile Views: 235+ Sponsor on GitHub License: MIT

🧭 How I Engineer AI Systems

  • Ground the output. Models only state what a tool or retrieved source supports, or escalate.
  • Bound the agent. Risk-classed tools, confirmed writes, tenant-scoped queries, fail-closed gates.
  • Measure, then claim. Every flagship ships tests, evals and a written list of its limits.

⚡ Featured Engineering & Flagship Architectures

🎮 World War B: Bugfront

Style-ranked 3D browser FPS, playable live.

  • Engine: Three.js PBR and Rapier3D WASM physics, under 22 MB.
  • Combat: 8 waves, a boss, 7 style ranks, 8 augment perks.
  • Duel mode: Supabase Realtime 1v1 with CPU fallback.

🎬 Pressflick

Free text match-cut video maker: one word holds, pages change.

  • Export: 1080p MP4 via WebCodecs, WebM fallback, 3 formats.
  • Private: No backend, account, watermark or tracking.
  • Verified: 111 E2E checks in 3 engines, zero axe issues.

🎙️ VoxFlow Voice Agent

Inbound voice agent for logistics desks, in supervised pilot.

  • Pipeline: Connect + Lex V2 → signed Lambda → FastAPI on Groq.
  • Tool safety: 22 identity-scoped tools, 2FA caller checks.
  • Evidence: 963 tests, a threat model, published limits.

🛡️ FinSentinel AML Engine

Local-first AML investigation with audit-logged cases.

  • Isolation: Postgres RLS and an append-only audit trail.
  • Speed: Evidence packs 1,070 ms → 44 ms p50, CI-gated.
  • Rigour: 771 tests, 29/29 release checks.

🏥 Clinical RAG Agent

Hypertension-care assistant with safety-first guardrails.

  • Retrieval: Dense + BM25 + rerank, orchestrated in LangGraph.
  • Calculators: Clinical maths in MCP tools, never the LLM.
  • Evaluated: 250+ tests and a 55-question golden set.

🔍 ExpertIQ Copilot

Expert discovery with vector search and knowledge graphs.

  • Pipeline: 6-node LangGraph RAG over ChromaDB and NetworkX.
  • 3D UI: Next.js 16 force-directed co-authorship graph.
  • Harness: Redis, Docker, RAGAS evals, 47 tests.

🧰 Guardrails for AI Coding Agents

Zero-telemetry local tools that stop leaks and false claims.

  • pre-push-verify: 30+ rules and entropy checks block leaks.
  • readme-guardian: PyPI/npm CLI that audits README facts.
  • Agent-native: Git hooks and Claude Code slash commands.

🧠 Cortex

Local-first second brain for Obsidian, with cited answers.

  • Retrieval: LanceDB + BM25 + wikilink graph, fused with RRF.
  • Privacy gate: Sensitive notes never reach unsafe providers.
  • Runtime: Thermal-aware indexing; exposes 5 MCP tools.

🚀 Additional Projects

Project Highlights Links
🎬 Shorts Autopilot (shorty) Operator-first YouTube Shorts pipeline: Groq scripts, FFmpeg renders, research gate, scheduled uploads. Live Demo • Repo
💸 Splittrack (expense-tracker) Multi-currency group expenses settled in the fewest transfers, on Next.js 14, Turso and Drizzle. Repo
🗺️ AI Engineering Roadmap Full-stack learning app with an SSE-streaming AI tutor, interview simulator and 5-phase roadmap. Repo

🛠️ Technical Stack

Agents & RAG   : LangGraph, LangChain, MCP, hybrid BM25 + RRF, ChromaDB, LanceDB
Voice          : Amazon Connect, Lex V2, Polly, Lambda, WebSockets, Supabase Realtime
Evals & Safety : LangSmith, RAGAS, golden sets, grounding guards, fail-closed gates
Backend & Data : Python, FastAPI, PostgreSQL (RLS), Redis, Supabase, Docker, Actions
Frontend & 3D  : TypeScript, Next.js, React 19, Tailwind, Three.js, WASM, WebCodecs
Testing        : Pytest, Vitest, Playwright, Ruff, mypy

💖 Support & GitHub Sponsors

Sponsor on GitHub

Sponsorship pays for evaluation compute, public demo hosting and open-source maintenance.

📈 Activity & Contribution Velocity

GitHub Statistics My Programming Languages
Contribution Graph
Built with precision by Jeevesh Singale • Open to AI engineering roles and agent architecture collaborations • Nottingham, UK

Pinned Loading

  1. pressflick pressflick Public

    Free text match-cut video maker for TikTok, Reels, Shorts and YouTube. Hold one word still, export 1080p MP4 in your browser. No watermark or signup.

    JavaScript

  2. WWB WWB Public

    World War B: Bugfront. High-octane, style-ranked 3D browser FPS built with Three.js, Rapier physics, and Supabase.

    TypeScript 2

  3. finsentinel-case-study finsentinel-case-study Public

    Public engineering case study for FinSentinel, an evidence-backed AML investigation assistant prototype

    Python 1

  4. clinical-rag-agent clinical-rag-agent Public

    Evidence-based clinical workflow assistant with agentic RAG for hypertension care with LangGraph orchestration, hybrid retrieval, OKF knowledge spine, safety-first guardrails, and Vercel deployment.

    Python 5

  5. pre-push-verify pre-push-verify Public

    The Autonomous Security & Anti-Slop Firewall for AI Coding Agents. Stop API key leaks, eliminate LLM code bloat, and verify tests before push.

    Python 1

  6. cortex cortex Public

    Building my second brain that will link to all my useful workflows

    Python 1