Shreyas Jagannath · Senior Applied AI Engineer, LondonProduction agents
built and owned end to end.
I give production agents the governed context they need, test them with evaluation gates, and build governance into the system. I wrote Nexus-MCP, an open-source code-intelligence server for AI coding agents, and the AI Hub that showed an engineering org what its AI spend was buying. I have audited AI systems against GDPR and the EU AI Act.
- Python
- TypeScript
- React
- FastAPI
- AWS
- GCP
- LangGraph
- MCP
Context
Nexus-MCP gives coding agents vector, BM25 and code-graph context. Published on PyPI, 607 tests.
See Nexus-MCPEvaluation
A golden dataset and scope checks test this site's chat agent. Run them before a release.
See the gatesGovernance
GDPR and EU AI Act audit at Move By Vision. Scoped agent skill for governed data at Lendable.
See the evidence
Shreyas Jagannath
Senior Applied AI Engineer · Context Engineering, Evaluation & Governance for Production Agents
Selected outcomes
- Fewer tokens on the heaviest agent messages, ~100k → ~10kKlyft · production
- 10x
- Faster core agent, 60s → 20sKlyft · production
- 3x
- Lower generation latency, 15s → 7sMove By Vision · production
- 55%
- Faster developer workflows, org-wideLendable · production
- 10%
- p95 latency of visual search with knowledge graphs, down from 15sAIMAGE · production
- 3–4s
- PyPI downloads of Nexus-MCPPyPI · pepy.tech
- 4,184
For recruiters
- Looking for
- Senior Applied AI Engineer
- Location
- London, UK (hybrid or on-site)
- Availability
- Immediate
About
Production software for environments where it has to work
From computer vision research (2016) to full-stack product engineering, enterprise consulting, founding-team builds, and cloud infrastructure — most recently AI and agentic systems, always with real users, measurable outcomes, and increasingly in regulated, safety-critical settings.
At Klyft, beyond building the core agentic workflows, I developed observability pipelines that surfaced an early signal: the generated plans were calibrated for high-performance users, while most of the actual user base were everyday consumers. Flagging this early helped the team adjust product direction before it became a retention problem. I enjoy this intersection of engineering and product thinking — using system behaviour and evaluation data to improve both architecture and outcomes.
Delivered a workshop at ODSC (Open Data Science Conference) on implementing AI and deep learning solutions in enterprise environments, to 100+ professionals. Earlier enterprise AI/ML proofs of concept for Airbus, Volvo and Target.
Diagnose
Align stakeholders on success criteria, constraints, risks, and what “good” looks like.
Architect
Choose the right trade-offs: guardrails, evaluation strategy, data contracts, and system boundaries.
Ship
Deliver production systems with instrumentation, reliability hardening, and iteration cycles.
The frontier timeline
A wave-by-wave narrative of shipped systems
The Agentic AI Engineer
Took an eight-agent coaching platform to production on Google ADK and Vertex AI. Cut tokens on the heaviest messages 10x (~100k → ~10k) by pre-summarising wearable data and scoping each agent's data access to its own tools. Evaluated Gemini 2.5 Flash against Pro on golden datasets to near-parity, so Pro never ran in production. Made the core workout-calibration agent 3x faster (60s → 20s) with async Pub/Sub processing.
Signal: Operating at the production edge of agentic AI. Applying the consulting diagnostic instinct to observability data to advise a go-to-market pivot.
The Reliability, Safety & Compliance Engineer
Inherited a Flutter + GCP platform in critical-failure alpha. Built a guard-railing and validation layer across Vertex AI integrations, removed 100% of AI crash failures over the release in alpha-stage testing, cut latency by 55% (15s to 7s), and conducted a GDPR/AI Act compliance audit.
Signal: AI governance done as engineering work: guardrails in code and a compliance audit, in a regulated healthcare setting. Also breadth of production AI experience—greenfield build, iterative product development, and rescue engineering.
The Independent Builder
Built independently between client work. Public: Nexus-MCP (on PyPI), Protash (live) and Health Intelligence Engine. Private, with walkthroughs on request: CR8, JobScout and Agent Harness.
Signal: Alongside client work, I build. Three of the systems are public and can be checked today: a package, a live product and a repo.
Agent context and governance in a regulated lender
Worked inside engineering teams to measure and improve how they actually used AI. Shipped on the organisation's Terraform/Terragrunt, GitHub Actions and Argo CD deployment stack, and set up Okta SCIM/OIDC authentication with admin/user group management for the tooling I built.
Signal: Chose a scoped, auditable agent skill over an MCP server for governed data access, to meet FCA security requirements.
The Researcher
Started not as a software engineer but as a researcher solving a genuinely hard spatial perception problem for a space mission. Mono vision depth detection — making a rover navigate a lunar environment using a single camera — is not a trivial problem. It requires understanding geometry, uncertainty, and edge cases.
Signal: Thinking about AI as a tool for solving hard real-world problems before most people were using the word 'AI' in their job titles. The origin of the computer vision seed.
The Builder
Immediately after research, went hands-on with production software building a food delivery platform, order processing, and bulk automation with Java, Spring, MySQL. Not glamorous — but important.
Signal: The full-stack foundation. Understanding systems end to end — databases, APIs, business logic, infrastructure. Java, Spring and MySQL in production; the layer beneath everything built since.
The Consultant
Delivering AI solutions across wildly different industries — automotive (TVS Motors), retail (Target), aerospace (Airbus), logistics (Volvo), fintech (Varthana), payments (Walmart/PhonePe). Delivered a workshop at ODSC (Open Data Science Conference) on implementing AI and deep learning in enterprise environments, training 100+ professionals.
Signal: Learned how AI lands — or fails to land — in real enterprise contexts. The diagnostic instinct starts here: 'what problem are you actually trying to solve?' before proposing a solution.
The Founder-Engineer
Built a company merging AI/AR virtual try-on engines and a custom graph-based vector-search architecture for visual search. Built custom vectoriser models and a graph vector database combining similarity search with graph traversal, with search and recommendations deployed to AWS on e-commerce platforms and the virtual try-on engine piloted on a retail client's storefront.
Signal: Leading people and building systems simultaneously. Translating a customer's floor-level problem into a technical architecture with real stakes.
The Founding Engineer
Owned the full technical direction. Built a Designer AI Assistant using RAG over design documents, and a manufacturing lifecycle platform. Researched Gaussian Splatting and PyTorch3D and made the call to integrate an API rather than build from scratch.
Signal: Technical roadmap ownership. Pragmatic trade-offs and shipping systems. Knowing when NOT to build is as valuable as knowing how to build.
Projects
Recent Work
Four projects, each labelled with how you can check it: public repo, live product, published package or private walkthrough.
Nexus-MCP: Hybrid Code Intelligence Server
- Live site
- Public repo
- Published package
Built to unblock a client engagement: a large, undocumented Flutter codebase that AI agents could not navigate reliably. Combines vector search, BM25 and a structural code graph (containment and imports), so an agent uses an estimated 70–90% fewer tokens to find relevant code and 30–60% fewer across a typical session. Published on PyPI (nexus-mcp-ci) with 3,500+ downloads and listed on the Glama MCP marketplace. 10 tools in one local server, tree-sitter parsing across 25+ languages, within a 350MB memory budget, no cloud dependency and no API keys.
Problem
AI coding agents spend most of their token budget reading irrelevant code — grepping for a function, reading entire files, repeating for every dependency. Existing code intelligence tools are fragmented across separate servers, each consuming 500 MB+ RAM.
Approach
Built a single MCP server consolidating two predecessor projects. Dual parsing (tree-sitter for symbol extraction, ast-grep for structural graph) feeds three search engines fused with Reciprocal Rank Fusion (vector 0.5 + BM25 0.3 + graph 0.2). Added an ONNX Runtime model path (~50 MB vs ~500 MB on PyTorch) alongside the default bge-small-en model, used LanceDB with mmap for disk-backed vectors, and added FlashRank re-ranking, token budgeting, semantic memory with TTL, and production hardening (graceful shutdown, corrupt index recovery, rate limiting, audit logging).
Result
10 tools within a 350 MB memory budget — hybrid search, symbol lookup, explain, an import and containment graph, persistent memory and architecture mapping across 25+ languages. Call-edge extraction, and the impact analysis that depends on it, is the open next step. 607 tests, 18 Architecture Decision Records, published on PyPI (pip install nexus-mcp-ci) with 3,500+ downloads and listed on the Glama MCP marketplace, CI/CD with trusted publishing. Zero API keys, fully local.
Health Intelligence Engine: Agentic Wellness Platform
- Public repo
Clinical AI platform that turns supplement queries into evidence-grounded, safety-checked recommendations. A 9-node LangGraph pipeline grounds product data against NIH DSLD and PubMed in a Neo4j knowledge graph, with a Pharmacovigilance Critic checking every recommendation against the user's allergies and medications.
Problem
Supplement and wellness recommendations lack grounding in clinical evidence and fail to account for individual contraindications, creating real safety risks for users.
Approach
9-node LangGraph pipeline grounding product data against NIH DSLD and PubMed in a Neo4j knowledge graph. A Pharmacovigilance Critic node checks every recommendation against the user's declared allergies and medications. DeepEval quality gates enforce output standards; published healf CLI exposes the platform.
Result
Evidence-grounded wellness guidance with built-in safety guardrails, contraindication checking, and DeepEval-validated output quality. Published as a CLI tool.
Protash: Autonomous Enterprise Prototyping Platform
- Live site
- Public repo
Describe a business intent; get a working, deployed prototype back in under 5 minutes. A hosted multi-agent tool that turns a described business goal into a running prototype autonomously — a 6-stage pipeline extracts domain context, spec and UX plan, generates a self-contained React component, then QAs and reviews it in parallel, self-healing through up to five correction rounds before anything reaches the screen.
Problem
Most prototyping tools produce generic wireframes — stakeholders see "Lorem ipsum" and "User A" and immediately check out, so a described business idea never gets a real gut-check before it's built.
Approach
Next.js/TypeScript multi-agent orchestrator and message bus: businessContext and spec agents extract domain entities, KPIs and BDD success criteria, a UX architect picks layout, charts and realistic domain-specific mock data, then a generator produces a sandboxed React component while QA and Reviewer agents run in parallel and request targeted fixes on failure. Built on the DeepSeek API to keep cost per run low, deployed on GCP Cloud Run via Cloud Build with runtime secret injection from Secret Manager, with Vitest, Playwright and an evaluation suite.
Result
A working, deployed prototype back from a plain-English business intent in under 5 minutes — wired with plausible domain data rather than placeholders.
CR8: Curriculum-to-Content Agent Pipeline
- Private, walkthrough on request
Private · walkthrough on request.
Built a 3-agent LangGraph pipeline (Ingest → Research → Generate) that turns raw university curriculum into a complete learning package: structured study guides, gap-analysis decks, video scripts, AI-narrated video, and a Bloom's-taxonomy MCQ quiz engine. A two-layer evaluation framework gates every output, and severity-based model routing keeps generation cost near zero without dropping quality where it matters.
Problem
Turning raw curriculum into structured learning content is manual, slow, and inconsistent — and without evaluation, generated educational material can't be trusted for real use.
Approach
3-agent LangGraph pipeline (Ingest → Research → Generate) producing study guides, gap-analysis decks, video scripts, AI-narrated video via Kokoro TTS, and Bloom's-taxonomy MCQ quizzes. A two-layer evaluation framework: free structural checks on every output, plus a DeepSeek-V3 LLM judge at roughly $0.02 per run, with severity-based routing across GPT-5-nano/mini/5.1.
Result
End-to-end curriculum-to-content system in production on Cloud Run — guides, decks, narrated video and assessment generated from a syllabus, with every output scored before it ships. 1,299 tests across pytest, Vitest and Playwright E2E, with zero real API calls in the suite.
More projects
Agent Harness: Secure Live-Data Dashboard Framework
- Private, walkthrough on request
A Python agent harness that connects to live data sources — databases, APIs, GCP infra — and builds validated dashboards, with every SQL path routed through one read-only guard by construction and no response model ever carrying a credential back to the LLM. In active development (MVP, Stage 3 of 5): the FastAPI gateway, SSE streaming and credential-hardening layer are built and tested (274 passing tests); the front-end and Cloud Run deploy are still ahead.
JobScout: Employer-Aware Job Discovery Pipeline
- Private, walkthrough on request
A daily pipeline that pulls roles from 13 ATS platforms and several job boards, checks each employer and salary against fixed rules, and scores what's left in two tiers: a free keyword pass first, then one LLM call only for jobs that survive it.
Context-Aware Knowledge Graph Agent
- Public repo
Built a Python-based NLP agent that converts raw consumer data into a Neo4j knowledge graph and structured vector store, feeding OpenAI, Gemini and Veo 3.1 video generation components with user-specific context for hyper-personalised video content.
Autonomous Agentic B2B Data CRM
- Public repo
Implemented an end-to-end agentic data pipeline to automatically crawl the web, qualify companies, enrich contacts and persist leads into an Enterprise GraphQL CRM.
Deployments & Architecture
Production Systems Shipped
From rescuing failing healthcare systems to cutting tokens on the heaviest agent messages 10x. Each card gives the result first and the detail under Show details.
Lendable Operations Ltd
Senior AI Engineer
Jul 2026 — Sep 2026
- Production system
Modelled engineering data into governed Snowflake metrics and gave engineers scoped, auditable agent access to it.
Klyft Technologies
Founding AI Engineer
Sep 2025 — Nov 2025
- Production system
Cut tokens on the heaviest agent messages 10x (~100k → ~10k) and made the core agent 3x faster (60s → 20s).
We Are Human Agency AB (Move By Vision)
Consultant (Pro Bono)
Dec 2025 — Jun 2026
- Production system
Cut generation latency 55% (15s → 7s) and audited the platform against GDPR and the EU AI Act.
How this was measured: 100% of AI-induced crash failures removed, measured over the release in alpha-stage testing.
Scan and Sew Inc.
Chief Data Officer (Founding Team)
Oct 2024 — Aug 2025
- Production system
Shipped a Designer AI Assistant (RAG over docs) and a real-time manufacturing tracking platform.
AIMAGE Technologies
Founder & CTO
Jan 2020 — Aug 2023
- Production system
Cut p95 visual-search latency from 15s to 3–4s with knowledge graphs, and grew the team to 10.
Mage Ventures Pvt. Ltd.
Software Consultant (Part-time)
Oct 2019 — Mar 2021
- Production system
Implemented a secure cross-platform payment gateway for Walmart-owned PhonePe across multiple client apps, and built an AI-powered automated grocery checkout engine using computer vision (image segmentation) for a product company.
Cellstrat Inc.
AI Consultant (Founding Team)
Oct 2018 — Oct 2020
- Production system
Consulted on AI/ML initiatives for enterprise clients, designing and implementing models in PyTorch, TensorFlow, Keras, and Python, containerised with Docker and deployed on AWS for their digital transformation journey.
EndGate Global
Full Stack Engineer (Founding Team)
Jul 2017 — Jul 2018
- Production system
Developed full-stack food delivery workflows that real businesses depend on, establishing fundamental production software engineering skills.
ISRO
AI Researcher Intern
Nov 2016 — Apr 2017
- Production system
Advanced the perception capabilities of the rover, learning the discipline of high-stakes, fault-intolerant research.
AI Governance
Safe, guardrailed and governed AI
I build AI systems that stay inside limits enforced in code: guardrails on model output, evaluation gates before release, access controls on data, and audits against GDPR and the EU AI Act. The work below runs in healthtech, wellbeing and enterprise settings, where a failure has a real cost.
Guardrailed agentic AI
Agents act only inside limits that the code enforces. A prompt is not the control.
- KlyftProduction systemScoped each agent's data access to its own tools, so no agent could read data outside its task.
- Move By VisionProduction systemBuilt a guard-railing and validation layer across the Vertex AI integrations of a healthtech platform and removed 100% of AI crash failures over the release in alpha-stage testing.
- Agent HarnessPrivate, walkthrough on requestRoutes every SQL path through one sqlglot read-only guard by construction, so an LLM agent with live database access cannot write.
- Health Intelligence EnginePublic repoA Pharmacovigilance Critic node checks every recommendation against the user's declared allergies and medications.
Evaluation gates and human oversight
Nothing ships on trust. Automated evaluation scores each output, and a person keeps the final say where the stakes are high.
- CR8Private, walkthrough on requestA two-layer evaluation framework gates every output: free structural checks first, then an LLM judge. Every output is scored before it ships.
- ProtashLive site · Public repoQA and Reviewer agents run in parallel and request fixes, with up to five correction rounds before a result reaches the screen.
- Health Intelligence EnginePublic repoDeepEval quality gates enforce output standards on the clinical recommendation pipeline.
- LendableProduction systemA manager-facing deny-list lets a person correct the automated PR-to-team attribution, so the model has a human correction loop.
- This siteLive siteThe JD Match analyzer must trace each claim to a profile entry and list unsupported requirements as gaps. The chat assistant has a scope lock, tested with npm run eval:scope. A recruiter agent may contact me only when a human asked it to.
Data safety and security
Secrets and personal data stay out of the model, the logs and the browser.
- Agent HarnessPrivate, walkthrough on requestNo response model carries a credential back to the LLM. Adversarial review closed a guard bypass and a ReDoS in the redaction regex (2.4s down to 8ms on a 64KB input).
- Nexus-MCPLive site · Public repo · Published packageRuns fully local with no cloud dependency and no API keys, so indexed code stays on the machine. Includes rate limiting and audit logging.
- ProtashLive site · Public repoGenerated React components run in a sandbox. Secrets reach Cloud Run at runtime from Secret Manager.
- This siteLive siteIP addresses are stored only as salted SHA-256 hashes. A per-request nonce CSP protects scripts. Chat sessions have owners, and Doppler holds the secrets.
Regulatory compliance
Compliance work is engineering work with an audit trail, done before release.
- Move By VisionProduction systemRan a full GDPR and EU AI Act compliance audit on a healthcare platform with strict compliance needs, to make the system releasable.
- This siteLive siteConsent tiers with opt-out, a privacy policy, and GDPR export and deletion endpoints that check IP-hash ownership.
Access control and governed data
The right people see the right data, and every metric has an owner.
- LendableProduction systemGave engineers agentic access to governed data through a scoped, auditable agent skill rather than an MCP server, to meet FCA security requirements.
- LendableProduction systemSet up Okta SCIM/OIDC authentication with admin and user groups for the tooling I built, on the organisation's own deployment stack.
- LendableProduction systemModelled incident.io, Jira, Datadog and GitHub data into governed, OKR-level Snowflake metrics, each tied to an owner and an outcome.
- LendableProduction systemBuilt the AI Hub, which shows how an engineering organisation uses AI across GitHub, Jira, Claude and Codex telemetry, so AI use and spend can be reviewed.
Skills
Technical Expertise
Core competencies spanning software engineering, cloud infrastructure, data platforms, and AI systems.
Context Engineering
Evaluation & Governance
Agents & Models
Engineering & Cloud
Data & Storage
Earlier work: MCP Servers & Code Intelligence, AI Coding Agent Tooling, AI Adoption Telemetry, Developer-Productivity Metrics, Workflow Analysis, Agent Skill Instrumentation & Optimisation, GitHub/Jira Data Pipelines, Metrics Dashboards, Claude Code & Cline Custom Harness (skills, agents, hooks, workflows), NLP, Computer Vision, Gaussian Splatting, PyTorch, TensorFlow, ONNX Runtime, Hugging Face.
Recognition
Awards, Certifications & Research
Most Innovative Business
Best Presentation Award
UKSEDS IOSM Competition Winner
Most Innovative Company
AI Hackathon Winner
Modern Machine Learning Approaches For Robotic Path Planning
Peer-Reviewed JournalEducation
Academic Background
MSc Artificial Intelligence
University of Surrey
2023 — 2024
- Dissertation on high-fidelity human avatar reconstruction using Gaussian Splatting for edge computing
- Co-delivered a local-infra LLM token-classification project that doubled render speed
- Winner of the most innovative business at UKSEDS IOSM Competition 2024 at Peryton Space Society
Master's in Computer Applications
Christ University
2013 — 2017
- Co-authored a peer-reviewed survey on AI path-planning methods
- ISRO collaboration: obstacle detection research for Chandrayaan Lunar Rover mission
- Built a full-stack Learning Management System with analytics
Bachelor's in Computer Applications
Bangalore University
2010 — 2013
- Built a Healthcare Management System to streamline patient care workflows
- Led the WEF Global Shapers team engaging C-level industry leaders on startup culture
Entrepreneurship / Entrepreneurial Studies
IFA Paris
- Winner of the Most Innovative Company award presented by IBM Paris and IFA Paris
- Legal, manufacturing, customer psychology, pricing and go-to-market strategy modules
Contact
Let's Connect
I'm currently open to Senior Applied AI, context engineering and AI governance roles — production agents, evaluation and governed data — where what I'm measured on is what I ship. I also consider forward deployed, solutions and developer-tooling roles. London, hybrid or on-site. Fastest route is email.