Fardin

Tech Lead - Platform & Engineering

Omnistra • Temps plein

Nov 2025 - Present • 11 mos

- Developed an ultra low-latency voice pipeline for real-time inbound and outbound calls, with barge-in, speculative turn-taking, and advanced audio processing, achieving sub-100ms responses over WebSockets and WebRTC. - Built an async streaming pipeline for high-concurrency environments, supporting thousands of simultaneous calls with minimal resource overhead. - Designed a node-based workflow engine for voice and chat, enabling dynamic routing, automation, and human handoff across sales and support flows. - Designed a tool-calling layer so agents execute external APIs and platform functions during live calls and chats, with secure third-party integrations. - Built omnichannel agents for WhatsApp, Messenger, Instagram, and an embeddable web widget, running the same agents and workflows as voice. - Developed Order OS for in-conversation catalog search, quoting, and order capture, with carrier routing, COD risk scoring, and automatic shipment booking. - Designed cross-channel customer identity resolution and a WooCommerce connector so buyer profiles, catalog, and orders stay consistent. - Developed knowledge-driven agent memory with async ingestion, hybrid retrieval, and scoped context injection, improving answer accuracy in conversation. - Built a multi-tenant SaaS with RBAC and org-level isolation, usage-based credit billing, and an approval-gated in-app operations copilot. - Developed scalable outbound campaigns with retry handling, increasing reliability and throughput for high-volume calling. - Integrated emotion-aware conversation handling with sentiment detection and response adaptation, plus analytics across voice, chat, and commerce. - Designed multi-region infrastructure with geo-routing, reducing call latency by up to 40%, and BYOC so clients use their preferred carriers and onboard faster.

Co-founder, Lead AI Software Engineer

IntellWe • Indépendant

Oct 2022 - Present • 4 yrs

- Built open-source multimodal ChatGPT alternative (text, image captioning, voice transcription), optimizing Hugging Face pipelines for real-time inference and expanding accessibility. - Created Python library for LLM token cost estimation, enabling developers to monitor and optimize budgets in production workflows, improving cost transparency. - Developed voice generator and transcription platform with multi-speaker diarization, real-time speech recognition, and AI-powered voice transformation for Wedio AI Studio. - Launched browser-based AI tool for one-click background removal, leveraging WebGPU acceleration for local processing, batch editing, and real-time previews. - Shipped privacy-first browser extension for context-aware text generation, backed by Next.js web app and FastAPI backend for authentication, billing, and LLM integrations. - Delivered GenAI solutions for 10+ clients across legal, retail, and enterprise domains, tailoring deployments to specific industry needs. - Built legal document summarization pipeline, reducing manual review time by 70% and increasing legal team productivity. - Deployed retail computer vision system (YOLOv7, TensorFlow), improving inventory accuracy by 90% and reducing stock discrepancies. - Prototyped GenAI MVPs using LangChain and GPT-3.5, shipping proof-of-concepts within 2 weeks to accelerate client innovation. - Reduced client cloud costs by 35% through TensorRT quantization and caching, optimizing resource utilization for AI workloads.

Lead AI/ML Engineer & Technical Recruiter

Prospect Engine LLC • Temps plein

Sep 2023 - Nov 2025 • 2 yrs 2 mos

- Designed modular agentic AI platform enabling creation of multi-agent systems with document ingestion, pgvector semantic search, and contextual query responses; authored official Python SDK with async and type hints. - Engineered outbound calling platform integrating Twilio, OpenAI Realtime API, and Deepgram for live speech-to-speech AI conversations, enabling workflow-driven call flows and analytics dashboard. - Architected and deployed local-machine-optimized LLM (RAG) with LangChain and Hugging Face, achieving 90x productivity gains by leading full-cycle development and performance tuning. - Built real-time meeting summarization and automated follow-up pipeline using WebSockets and OpenAI, resulting in 95%+ user satisfaction and streamlined meeting workflows. - Developed resume auto-classification (95% accuracy) and ranking system using FAISS similarity search, scaling to process 5,000 resumes per hour for recruitment automation. - Enhanced LinkedIn automation system by building scraping engine with Puppeteer and proxy APIs, adding payment flows and geo-targeted proxy management for increased automation efficiency. - Built cross-platform messaging hub unifying Gmail, Slack, WhatsApp, and Notion into one inbox, implementing daily summaries and OAuth-based APIs for seamless productivity.