Jayachandran Ramadoss profile photo 1 on Tal

Jayachandran Ramadoss

Lead Software Engineer @ DevRev | Core AI · Multi-Agent Systems · Agentic RAG · Data & Analytics

Chennai

4+ years of work experience

DevRev
Jul 20264 mosPresent

Chennai, Tamil Nadu, India

I build the AI and data platform behind DevRev. I turn core AI and large-scale data into systems that ship and hold up in production. Lead engineer for Agents & Analytics — the multi-agent runtime (orchestration, agentic RAG, core AI) on one side, and a real-time, multi-tenant analytics platform on the other: CDC pipelines, Temporal ETL, and edge DuckDB handling millions of tickets, issues, and conversations across 21,000+ organizations and 6 regions.
DevRev
Jul 20263 moPresent

Chennai

Software Engineer, Data Analytics | December 2025–June 2026 Technical lead across two teams spanning agents, analytics, data foundations and workflows. On-device AI and model systems — Offline Edge Computer First-place hackathon team project; direct co-founder endorsement and equity award. • Architected a custom on-device agent runtime with direct llama.cpp integration, quantized GGUF models, Metal GPU inference and Electron process isolation; implemented an Anthropic-compatible API while keeping LLM inference and prepared-context retrieval local. • Engineered small-model safeguards through serialized inference, context-window budgeting and JSON Schema-constrained decoding for structured tool calling, with token-budgeted transcript trimming and validated tool selection. • Built on-device retrieval-augmented generation (RAG) using dense embeddings, SQLite FTS5/BM25 and reciprocal rank fusion (RRF), grounding responses in conversation-scoped local context. • Designed a cloud-to-edge model-adaptation pipeline combining teacher-trace validation, knowledge distillation and LoRA-based parameter-efficient fine-tuning (PEFT) for compact local models. Autonomous agent systems and orchestration • Architected and engineered a multi-agent ticket-triage system used across 3,000+ tickets, orchestrating specialist agents for knowledge retrieval, runtime diagnostics, source-code investigation and engineering handoffs. • Designed evidence-gated orchestration with independent LLM critique, bounded investigation budgets, delta-aware follow-ups and controlled lesson capture, keeping customer responses as human-reviewed drafts. Search, agent runtimes and workflow engineering • Led feature engineering and implementation for OpenSearch-backed Wisp hybrid search, spanning KNN/vector semantic retrieval, BM25 lexical search, weighted RRF and reranking with rank-fusion fallback. • Engineered agent SDK and Agent Studio capabilities for tool calling, prompt guidance, orchestration
DevRev
Dec 20258 mosJul 2026

Chennai, Tamil Nadu, India

Senior engineer at the intersection of large-scale data infrastructure, edge query engines, and agentic AI. On DevRev's Data Analytics team I worked end-to-end across the whole platform; the AI-agent and edge-inference work I drove in parallel became my Lead charter (Jul 2026). Data platform — end-to-end: - Engineered in this pipeline top to bottom: MongoDB CDC → Kafka → BigQuery → Temporal ETL → Parquet → catalog → DuckDB engines in cloud, browser (WASM), and on-device edge. - In multi-tenant SaaS serving millions of tickets, issues, and conversations across 6 regions, I hardened and debugged what broke: led the Edge Compute engine migration; - Cut edge idle memory 12× via allocator-level profiling + adaptive management; tuned real-time incremental sync + schema evolution; hardened network resilience (NetworkGuard, mTLS) AI, agents & inference: - I architected autonomous multi-agent ticket triage — plan-execute-synthesize orchestration over live telemetry, with confidence scoring, adversarial audits, and MCP tool offload — and contributed across the wider ecosystem (KYC, Lighthouse, IMA) plus agent skills and RAG workflows used by teammates in production. - Where I go deepest is edge AI: I connect retrieval, memory, and inference into one local-first system — GraphRAG/Graphify retrieval, self-improving memory with reinforcement loops, LoRA adaptation, knowledge distillation, and quantized edge models running NL-to-SQL fully offline on no-GPU laptops. Its 4-tier fallback is the core contract, not a safety net: DistilBERT intent → RoBERTa schema linking → constrained decoding (SQL FSM + schema trie) that prunes the output space so the model degrades gracefully instead of hallucinating. Won 1st prize solo at DevRev's hackathon. Stack: Go · TypeScript · Python · MongoDB · Kafka · BigQuery · Temporal · DuckDB · WASM · GraphRAG · LoRA · Knowledge Distillation · Constrained Decoding · Edge Inference · MCP
Jul 20222 yrs 7 mos2025

Recognition: Awesome Performer (2024); Passionate Programmer (2023). Applied ML and predictive features • Engineered NLP features for field prediction, sentiment analysis and text classification; trained and tuned SBERT-based field-prediction classifiers using historical records and related-field context. • Designed configurable ML training and inference pipelines across portals, modules and fields, covering text preprocessing, vectorization, model-to-field mapping, scheduled retraining, prediction APIs and user-feedback capture. • Benchmarked concurrent model training and inference across portals; profiled CPU/memory behavior, diagnosed model-reload stalls and tuned caching in load-test environments. Core product and systems engineering • Redesigned Java mail processing as a producer-consumer pipeline, separating single-threaded fetching, database-backed staging and parallel workers for new-mail, conversation and AI workloads. • Led PostgreSQL upgrade and reliability work across ServiceDesk Plus and AssetExplorer, covering automated migration, backup/restore safeguards, compatibility fixes and security-driven updates. • Developed a configurable business-module builder for extending ServiceDesk Plus with custom application modules. Engineering automation • Built a configuration-driven Java framework to install, start, restore and upgrade product instances across Windows and Linux, coordinating remote execution over WinRM and SSH with centralized status reporting. AI integration • Built a Java-based Model Context Protocol (MCP) server to expose product capabilities through a standardized interface for AI clients and tool invocation.
Zoho
May 20223 yrs 8 mosDec 2025

Chennai, Tamil Nadu, India

Backend & platform engineer on ManageEngine ServiceDesk Plus — a flagship enterprise ITSM product used by thousands of customers under strict SLAs. I owned and shipped backend, database, and AI-enabled modules across the product. What I worked on: • AI-enabled backend: designed SBERT-based field prediction across 36 languages (self-improving retraining loop, Llama-3/Qwen synthetic data) and a standards-compliant MCP server (Java + Python, OpenAPI, provider-neutral adapters for OpenAI/Claude/internal models) — plus GPT-based generation, classification, and summarization with hallucination controls. • Distributed & concurrent systems: designed and improved high-throughput mail/messaging pipelines (POP, IMAP, EWS, Microsoft Graph) with thread-pool isolation, retries, and backpressure. • Databases & reliability: owned bundled PostgreSQL end-to-end — zero-downtime upgrades, performance tuning, security hardening, and live corruption recovery (pg_toast/FSM/VM) on customer data. • Identity & security: designed Login and OAuth2/JWT flows balancing security, backward compatibility, and UX. • Platform frameworks & UI: built a custom ORM/mapper, caching, and REST/OpenAPI frameworks; delivered Ember.js (HBS), JavaScript, HTML/CSS components wired to backend and AI workflows. Impact: • ~180% higher throughput in critical mail workflows; 40% faster queries via caching + tuning. • ~95% fewer recurring support issues on owned modules; 98% support rating. • ~75% shorter test/release cycles through platform automation and tooling. • Two internal awards — Awesome Performer (2024) and Passionate Programmer (2023) — for ownership of critical, customer-impacting systems. Core domains: Backend & Platform · Distributed Systems · AI/LLM Integration · PostgreSQL · Identity/OAuth · Enterprise SaaS Stack: Java · Spring · PostgreSQL · Python · REST · OpenAPI · Microservices · Java Concurrency · SBERT · ONNX · MCP · Ember.js · OAuth2 · JWT
Zoho
May 20212 mosJun 2021

Chennai, Tamil Nadu, India

- Product Automation Tool – Java-Based Admin–Client Automation Framework (Internship Project) Tech Stack: Java 11, SSH, WinRM, JSON Configurations, Multi-OS Automation (Windows, Ubuntu, RedHat, CentOS), CI/CD Concepts. - Built an Admin–Client automation framework to remotely install, upgrade, restore backups, and manage product instances across multiple OS environments using native protocols—no external tools required. - Designed one-to-many admin–client architecture, with secure authentication and OS-native execution (SSH/WinRM). - Implemented process tracking, cloud file handling, and email reporting for complete visibility of operations. - Impact: Reduced testing effort by 75%, accelerated feature releases across environments, and automated repetitive developer workflows.

Jayachandran’s AI Stack

claude code
claude code
codex
codex
kimi
kimi
notebooklm
notebooklm
obsidian
obsidian
opencode
opencode

Education

Sri sairam engineering college

Bachelor of Engineering, Electronics and Communications Engineering

Sri sairam engineering college

20222022

Join Jayachandran and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Jayachandran Ramadoss on Tal

Headline

Lead Software Engineer @ DevRev | Core AI · Multi-Agent Systems · Agentic RAG · Data & Analytics

About

I build AI systems and data platform behind DevRev that survive production — autonomous agents, edge inference, and the large-scale data platforms are underneath them. Lead Software Engineer at DevRev (Agents & Analytics). On the AI side I architect production agentic systems — multi-agent triage (plan-execute-synthesize, confidence scoring, adversarial audits, MCP tool offload) and RAG workflows shipping in internal triage. I work end-to-end across the data path — MongoDB CDC, Kafka, BigQuery, Temporal ETL, DuckDB across cloud, WASM, and on-device edge — hardening multi-tenant SaaS that serves millions of tickets, issues, and conversations. The rare part is the combination — cloud data platform, edge device, and the AI layer between. I don't think AI replaces great engineers; it raises the bar for what they can do.

Location

Chennai

AI stack

claude code

codex

kimi

notebooklm

obsidian

opencode

Work experience

Lead Software Engineer at DevRev, Jul 2026 - Present · 4 mos, Chennai, Tamil Nadu, India, I build the AI and data platform behind DevRev. I turn core AI and large-scale data into systems that ship and hold up in production. Lead engineer for Agents & Analytics — the multi-agent runtime (orchestration, agentic RAG, core AI) on one side, and a real-time, multi-tenant analytics platform on the other: CDC pipelines, Temporal ETL, and edge DuckDB handling millions of tickets, issues, and conversations across 21,000+ organizations and 6 regions.

Software Engineer at DevRev, Jul 2026 - Present · 3 mo, Chennai, Software Engineer, Data Analytics | December 2025–June 2026 Technical lead across two teams spanning agents, analytics, data foundations and workflows. On-device AI and model systems — Offline Edge Computer First-place hackathon team project; direct co-founder endorsement and equity award. • Architected a custom on-device agent runtime with direct llama.cpp integration, quantized GGUF models, Metal GPU inference and Electron process isolation; implemented an Anthropic-compatible API while keeping LLM inference and prepared-context retrieval local. • Engineered small-model safeguards through serialized inference, context-window budgeting and JSON Schema-constrained decoding for structured tool calling, with token-budgeted transcript trimming and validated tool selection. • Built on-device retrieval-augmented generation (RAG) using dense embeddings, SQLite FTS5/BM25 and reciprocal rank fusion (RRF), grounding responses in conversation-scoped local context. • Designed a cloud-to-edge model-adaptation pipeline combining teacher-trace validation, knowledge distillation and LoRA-based parameter-efficient fine-tuning (PEFT) for compact local models. Autonomous agent systems and orchestration • Architected and engineered a multi-agent ticket-triage system used across 3,000+ tickets, orchestrating specialist agents for knowledge retrieval, runtime diagnostics, source-code investigation and engineering handoffs. • Designed evidence-gated orchestration with independent LLM critique, bounded investigation budgets, delta-aware follow-ups and controlled lesson capture, keeping customer responses as human-reviewed drafts. Search, agent runtimes and workflow engineering • Led feature engineering and implementation for OpenSearch-backed Wisp hybrid search, spanning KNN/vector semantic retrieval, BM25 lexical search, weighted RRF and reranking with rank-fusion fallback. • Engineered agent SDK and Agent Studio capabilities for tool calling, prompt guidance, orchestration

Software Engineer at DevRev, Dec 2025 - Jul 2026 · 8 mos, Chennai, Tamil Nadu, India, Senior engineer at the intersection of large-scale data infrastructure, edge query engines, and agentic AI. On DevRev's Data Analytics team I worked end-to-end across the whole platform; the AI-agent and edge-inference work I drove in parallel became my Lead charter (Jul 2026). Data platform — end-to-end: - Engineered in this pipeline top to bottom: MongoDB CDC → Kafka → BigQuery → Temporal ETL → Parquet → catalog → DuckDB engines in cloud, browser (WASM), and on-device edge. - In multi-tenant SaaS serving millions of tickets, issues, and conversations across 6 regions, I hardened and debugged what broke: led the Edge Compute engine migration; - Cut edge idle memory 12× via allocator-level profiling + adaptive management; tuned real-time incremental sync + schema evolution; hardened network resilience (NetworkGuard, mTLS) AI, agents & inference: - I architected autonomous multi-agent ticket triage — plan-execute-synthesize orchestration over live telemetry, with confidence scoring, adversarial audits, and MCP tool offload — and contributed across the wider ecosystem (KYC, Lighthouse, IMA) plus agent skills and RAG workflows used by teammates in production. - Where I go deepest is edge AI: I connect retrieval, memory, and inference into one local-first system — GraphRAG/Graphify retrieval, self-improving memory with reinforcement loops, LoRA adaptation, knowledge distillation, and quantized edge models running NL-to-SQL fully offline on no-GPU laptops. Its 4-tier fallback is the core contract, not a safety net: DistilBERT intent → RoBERTa schema linking → constrained decoding (SQL FSM + schema trie) that prunes the output space so the model degrades gracefully instead of hallucinating. Won 1st prize solo at DevRev's hackathon. Stack: Go · TypeScript · Python · MongoDB · Kafka · BigQuery · Temporal · DuckDB · WASM · GraphRAG · LoRA · Knowledge Distillation · Constrained Decoding · Edge Inference · MCP

Software Engineer at ZOHO CORPORATION — MANAGEENGINE, Jul 2022 - 2025 · 2 yrs 7 mos, Recognition: Awesome Performer (2024); Passionate Programmer (2023). Applied ML and predictive features • Engineered NLP features for field prediction, sentiment analysis and text classification; trained and tuned SBERT-based field-prediction classifiers using historical records and related-field context. • Designed configurable ML training and inference pipelines across portals, modules and fields, covering text preprocessing, vectorization, model-to-field mapping, scheduled retraining, prediction APIs and user-feedback capture. • Benchmarked concurrent model training and inference across portals; profiled CPU/memory behavior, diagnosed model-reload stalls and tuned caching in load-test environments. Core product and systems engineering • Redesigned Java mail processing as a producer-consumer pipeline, separating single-threaded fetching, database-backed staging and parallel workers for new-mail, conversation and AI workloads. • Led PostgreSQL upgrade and reliability work across ServiceDesk Plus and AssetExplorer, covering automated migration, backup/restore safeguards, compatibility fixes and security-driven updates. • Developed a configurable business-module builder for extending ServiceDesk Plus with custom application modules. Engineering automation • Built a configuration-driven Java framework to install, start, restore and upgrade product instances across Windows and Linux, coordinating remote execution over WinRM and SSH with centralized status reporting. AI integration • Built a Java-based Model Context Protocol (MCP) server to expose product capabilities through a standardized interface for AI clients and tool invocation.

Member Technical Staff at Zoho, May 2022 - Dec 2025 · 3 yrs 8 mos, Chennai, Tamil Nadu, India, Backend & platform engineer on ManageEngine ServiceDesk Plus — a flagship enterprise ITSM product used by thousands of customers under strict SLAs. I owned and shipped backend, database, and AI-enabled modules across the product. What I worked on: • AI-enabled backend: designed SBERT-based field prediction across 36 languages (self-improving retraining loop, Llama-3/Qwen synthetic data) and a standards-compliant MCP server (Java + Python, OpenAPI, provider-neutral adapters for OpenAI/Claude/internal models) — plus GPT-based generation, classification, and summarization with hallucination controls. • Distributed & concurrent systems: designed and improved high-throughput mail/messaging pipelines (POP, IMAP, EWS, Microsoft Graph) with thread-pool isolation, retries, and backpressure. • Databases & reliability: owned bundled PostgreSQL end-to-end — zero-downtime upgrades, performance tuning, security hardening, and live corruption recovery (pg_toast/FSM/VM) on customer data. • Identity & security: designed Login and OAuth2/JWT flows balancing security, backward compatibility, and UX. • Platform frameworks & UI: built a custom ORM/mapper, caching, and REST/OpenAPI frameworks; delivered Ember.js (HBS), JavaScript, HTML/CSS components wired to backend and AI workflows. Impact: • ~180% higher throughput in critical mail workflows; 40% faster queries via caching + tuning. • ~95% fewer recurring support issues on owned modules; 98% support rating. • ~75% shorter test/release cycles through platform automation and tooling. • Two internal awards — Awesome Performer (2024) and Passionate Programmer (2023) — for ownership of critical, customer-impacting systems. Core domains: Backend & Platform · Distributed Systems · AI/LLM Integration · PostgreSQL · Identity/OAuth · Enterprise SaaS Stack: Java · Spring · PostgreSQL · Python · REST · OpenAPI · Microservices · Java Concurrency · SBERT · ONNX · MCP · Ember.js · OAuth2 · JWT

Summer Intern at Zoho, May 2021 - Jun 2021 · 2 mos, Chennai, Tamil Nadu, India, - Product Automation Tool – Java-Based Admin–Client Automation Framework (Internship Project) Tech Stack: Java 11, SSH, WinRM, JSON Configurations, Multi-OS Automation (Windows, Ubuntu, RedHat, CentOS), CI/CD Concepts. - Built an Admin–Client automation framework to remotely install, upgrade, restore backups, and manage product instances across multiple OS environments using native protocols—no external tools required. - Designed one-to-many admin–client architecture, with secure authentication and OS-native execution (SSH/WinRM). - Implemented process tracking, cloud file handling, and email reporting for complete visibility of operations. - Impact: Reduced testing effort by 75%, accelerated feature releases across environments, and automated repetitive developer workflows.

Education

Bachelor of Engineering, Electronics and Communications Engineering, Sri sairam engineering college