Jeevesh Singale jeevesh2515

Jeevesh Singale

AI & Systems Engineer | Voice AI Systems, Agentic Orchestration & 3D WebGL

Building Low-Latency Voice Agents (<200ms), Zero-Hallucination Clinical RAG, Local-First AML Engines, and In-Browser 3D Systems.

London, UK • LinkedIn • Portfolio • Email • Sponsor

⚡ Featured Engineering & Flagship Architectures

🎮 World War B: Bugfront

High-Octane, Style-Ranked 3D Browser FPS Built with Pure WebGL & WASM.

WASM Physics & Engine: Three.js PBR rendering + Rapier3D WASM physics running at locked 60 FPS in-browser with zero install
Style-Ranked Combat: 8 escalating waves, Devil May Cry-inspired style ranks (D → FRONTBREAKER), 8 stackable tactical augment perks, and Overdrive state
Cloud Leaderboard: Supabase real-time cloud profile sync, telemetry, and global competitive leaderboard

⚡ VoxFlow Voice Agent

Bilingual Conversational Voice AI Platform for Supply Chain & FMCG (Commercial SaaS).

Turn Latency: Sub-200ms glass-to-glass turn latency (Amazon Connect → Groq Whisper STT → Llama 3 70B → Edge TTS)
Streaming Stack: Raw PCM WebSockets + Silero VAD + speculative interruptible barge-in under heavy network jitter
Stateful DAG: Live warehouse stock validation & dynamic PO dispatch with LangGraph; 952 passing tests

🛡️ FinSentinel AML Investigation Engine

Local-First AML Triage & Evidence Synthesis SaaS (Private Enterprise / Client Pilot).

Deterministic Triage: Tenant-scoped transaction ingestion, leakage-free feature extraction & scikit-learn IsolationForest anomaly scoring
Human-in-the-Loop SARs: LangGraph case review workflow with regulatory citation grounding, zero-hallucination guardrails, and audit trails
Rigorously Verified: 10 completed phases (P1–P10), 371 tests passing (285 unit + 86 integration), HMAC licensing, and client-side SHA-256 evidence hashing

🏥 Clinical RAG Workflows

Evidence-Based Clinical Workflow Assistant with Zero-Hallucination Guardrails.

Zero-Hallucination Spine: Medical knowledge stored in Google Open Knowledge Format (OKF) for deterministic biomedical RAG
Deterministic Calculators: Native biomarker mathematical engines (eGFR, MAP, pulse pressure) + edge refusal safety filters
Clinical Benchmarks: Verified with 258 passing tests + 55 golden clinical test cases evaluated on LangSmith and RAGAS

🔍 ExpertIQ Copilot

Enterprise Research Intelligence & Expert Discovery Platform.

Hybrid RAG & Knowledge Graph: 6-node LangGraph pipeline orchestrating ChromaDB dense vectors + NetworkX co-authorship graphs with HyDE
Force-Directed 3D UI: Next.js 16 + React 19 interactive frontend with dynamic D3 3D graph exploration and author influence ranking
Production Architecture: Redis caching layer, Dockerized container deployment, and 47 automated end-to-end tests

🛡️ Pre-Push Verify & Developer Tooling

Autonomous Security Firewall & Guardrails for AI Coding Agents.

Pre-Push Verify: Stops API key leaks, eliminates speculative LLM code slop via 5 Ponytail pruning laws, and verifies test suites before commit/push
Readme Guardian: Documentation freshness engine distributed via custom Homebrew Tap (brew install jeevesh2515/tap/readme-guardian)
Agent-Ready: Air-gapped, zero telemetry, 100% test coverage; built for Claude Code, Antigravity, Cursor, and Windsurf workflows

🚀 Additional Open-Source Projects & Specialized Systems

Project	Highlights & Architecture	Links & Status
🎬 Shorts Autopilot (`shorty`)	Low-cost, operator-first YouTube Shorts video automation pipeline. Uses Groq Llama 3 70B for scripting, YouTube Data API v3 for topic discovery, and local/Docker FFmpeg for rendering.	Live Demo • Repo
💸 Splittrack (`expense-tracker`)	Multi-currency group expense tracker & debt settlement engine. Computes minimal peer-to-peer transfer transactions. Built with Next.js 14 App Router, Turso libSQL, Drizzle ORM, and NextAuth.	Repo
🧠 Cortex (`cortex`)	Local-first cognitive memory engine and knowledge graph accelerator designed to link notes, developer workflows, and autonomous agent contexts.	Repo
🗺️ AI Engineering Roadmap	Community roadmap application curating top-tier free AI engineering courses, tutorials, and frameworks into a structured progression dashboard.	Live App • Repo

🛠️ Technical Stack & Engineering Capabilities

Core AI & Agents   : LangGraph (Stateful DAGs), LlamaIndex, Semantic Kernel, CrewAI, AutoGen, HyDE
Voice & Realtime   : Amazon Connect, Lex V2, Silero VAD, Groq Whisper STT, Edge TTS, WebSockets, Audio Worklets
3D WebGL & Physics : Three.js (PBR Materials, Shaders), Rapier3D (Rust WASM Physics), WebGL, D3.js, Howler.js
RAG & Knowledge    : Hybrid (Dense + BM25), GraphRAG, Google OKF, ChromaDB, LanceDB, pgvector, NetworkX
Backend & Data     : Python 3.11+, FastAPI, PostgreSQL 16 (RLS, Alembic), Turso (libSQL), Supabase, Docker, Redis
Frontend & Web     : Next.js 16/15, React 19, TypeScript (Strict), TailwindCSS v4, Vite, Web Crypto API
Safety, Evals & CI : LangSmith, Ragas, DeepEval, Deterministic Refusals, STRIDE Threat Modeling, Pytest, Ruff

💖 Support & GitHub Sponsors

I actively build zero-hallucination medical AI systems, high-speed voice streaming agents, WASM 3D engines, and open-source developer guardrails. Your sponsorship funds cloud evaluation compute, biomedical benchmarking infrastructure, and continuous open-source maintenance.

Tier	Focus	Impact
Supporter & Backer	Open-source stewardship	Backs open research datasets and public web demo hosting
Startup & Tooling Sponsor	Developer ecosystems	Sponsor logo and link featured on project READMEs & documentation
Corporate Advisory Partner	Architecture consulting	1-on-1 technical advisory on low-latency voice, RAG guardrails & agent DAGs

Become a GitHub Sponsor →

📈 Activity & Contribution Velocity

_{Engineered with precision by Jeevesh Singale • Open to AI Engineering & Agent Architecture collaborations • London, UK}

Navigation Menu