# Sairam Ugge — GenAI & Backend Engineer @ Ascendion (Jun 2024 – Present)
Location: Hyderabad, India
Summary: I build production multi-agent LLM systems and the event-driven backends behind them. I co-built Pensieve and architected the AAVA Code backend at Ascendion, and build open-source AI infrastructure in the open.
Links: GitHub https://github.com/sairam0424 · LinkedIn https://linkedin.com/in/sairam0424 · npm https://www.npmjs.com/~sairamugge-0000 · PyPI https://pypi.org/user/Sairam0000/ · Dev.to https://dev.to/sai_ram_0000 · Substack https://sairam0000.substack.com
Contact: uggesairam0000@gmail.com

## Production Work (at Ascendion)
### Pensieve (AI Process-Orchestration Engine · Ascendion)
Contribution: Co-built · production-hardened
I co-built a multi-agent LLM orchestration engine with human-in-the-loop approval gates, real-time streaming, and governed LLM routing across cloud providers.
Metrics: 2K+ daily users across domains; HITL approval gates; Multi-cloud governed LLM routing
Tech: Python, LLM Orchestration, Multi-Agent, SSE, Redis Streams, Cloud Routing

### AAVA Code (AI Coding Plugin for VS Code · Ascendion)
Contribution: Co-built · architected the backend
I co-built a single-agent prototype's re-architecture into a multi-agent orchestration system (Main-Agent + sub-agent via crewAI flows) powering an AI coding plugin for VS Code.
Metrics: 3K+ daily users; 5+ client environments; 150+ skills · 40+ tools · ~60 commands
Tech: Python, crewAI, Multi-Agent, VS Code, Backend Architecture

### Wireframe Generator (GenAI Wireframe Generator · Ascendion)
Contribution: Co-built · production-ready
I co-built a GenAI-powered wireframe generator that converts product inputs (PRDs, sketches, prompts) into production-ready UI artifacts, delivered as a full vertical slice within Experience Studio over an SSE real-time architecture.
Metrics: 40% fewer UX iteration cycles (5 rounds → 3); 60% faster prototyping; 500+ users daily
Tech: GenAI, Angular, SSE, Real-Time Rendering, Backend Stream Handling

### Prompt-to-React (Prompt-to-React Code Generation · Ascendion)
Contribution: Co-built · production-ready
I co-built a prompt-to-React code generation system that turns generated wireframes into production-ready modular components and routing logic.
Metrics: 50% less frontend dev time (2 days → 1 per feature); 55% less manual coding effort; 2K+ developers; 50+ teams
Tech: React, Code Generation, Modular Components, Routing Logic, Wireframe Input

### Execution Engine (Prompt-Driven Execution Engine · Ascendion)
Contribution: Co-built
I co-built a prompt-driven execution engine using LLM-based agent orchestration with RAG and ReAct pipelines that automates generation of structured product artifacts.
Metrics: 1.5h → 15m workflow planning time; 65% → 85% first-pass acceptance; 1.5K+ users
Tech: LLM Agent Orchestration, RAG, ReAct, Prompt-Driven

## Open-Source Projects (github.com/sairam0424)
### Kelvran — Self-hosted Go LLM gateway with a correctness-gated cache and per-run cost attribution
Group: Platform, Guardrails & Observability · 863 commits
Repo: https://github.com/kelvran/gateway
Self-hosted LLM gateway in Go that speaks OpenAI's Chat Completions and Embeddings wire format, plus Anthropic's Messages API, in front of OpenAI, Anthropic, Gemini, AWS Bedrock and any OpenAI-compatible server. One binary carries a three-layer response cache whose near-duplicate hits must pass non-configurable correctness gates, per-tenant virtual keys with Decimal budgets and Redis-backed rate limits, regex guardrails, and OpenTelemetry spans that attribute cost to an agent_run_id. A separate Python CLI, evals, scores agent outputs with Wilson intervals and an opt-in two-judge panel. 23 gateway releases through v0.20.0 (2026-10-15); solo-maintained.
Tech: Go, Python, Redis, bbolt, OpenTelemetry, Protobuf, AWS Bedrock, Docker, Kubernetes, Terraform

### toolwarrant — Approve one exact MCP tool-definition digest; the gateway lists only tools that match it
Group: Platform, Guardrails & Observability · 445 commits
Repo: https://github.com/sairam0424/toolwarrant
Go control plane against MCP tool poisoning by rug pull. It crawls each upstream MCP server, versions every tool definition by the SHA-256 of its RFC 8785 canonical JSON (`_meta` included), and lets a named person approve one exact digest. It runs as agentgateway v1.6.0's ExtMCP processor, so a tool is listed only while its live digest still equals the approved one, and a call is denied whenever the evidence is missing or disagrees. Every decision lands in a hash-chained SQLite audit log, and since v0.4 human decisions are signed with OpenSSH keys. Single-tenant, CLI-only, pinned to one gateway version, one author.
Tech: Go, gRPC, SQLite, MCP, agentgateway, RFC 8785 JCS, OpenSSH sshsig, Sigstore cosign, SLSA

### Cost Guard MCP — Fail-closed query-cost guardrails for AI agents on BigQuery, Snowflake and Databricks
Group: Platform, Guardrails & Observability · 211 commits
Repo: https://github.com/mcpsmiths/cost-guard-mcp
A single-process stdio MCP server that prices a SQL statement before an AI agent runs it: BigQuery dry run (PRECISE), Snowflake EXPLAIN (UPPER_BOUND), Databricks EXPLAIN COST (HEURISTIC). It refuses to execute when the estimate exceeds the caller's cap or when no comparable figure exists. Four tools, a schema-enforced accuracy tier on every estimate, read statements only by default, and results capped at 1000 rows and 8 MiB and labelled untrusted data. Ten PyPI releases in 37 days, the last three remediating a security audit, with SLSA provenance and a hash-pinned build. Solo-maintained; one GitHub star.
Tech: Python, MCP, Pydantic, BigQuery, Snowflake, Databricks, uv, pytest, OpenTelemetry, Docker

### prompt-lift — Claude Code mod: one press rewrites a vague prompt from your CLAUDE.md and skills
Group: Tooling & Lab · 178 commits
Repo: https://github.com/sairam0424/claude-prompt-lift
A Claude Code mod, an in-process hooks module rather than a skill, that draws an Enhance / Send / Undo band above the prompt. One press sends the draft, the instruction files the session already loaded and the skill listing to a three-rung model chain with per-rung fallback, highlights the inserted words, and prints the call's cost on the band. Nine post-call guards refuse rewrites that add commands, invent paths or flip intent, and optional session and day spend caps sit on a per-writer ledger. MIT, 178 commits at v0.6.0, installed from GitHub as a plugin marketplace; not on npm.
Tech: TypeScript, Claude Code Mods API, Anthropic Models API, Prompt Caching, AWS Bedrock, Bash, tmux, asciinema

### RateCap — Recreation of Stripe's four-tier rate limiter and load shedder, as a Go sidecar and core
Group: Platform, Guardrails & Observability · 347 commits
Repo: https://github.com/sairam0424/RateCap
Open-source recreation of the four-tier rate limiter and load shedder Stripe described in a 2017 engineering post: a per-key token bucket, a per-key concurrent-request cap, a fleet-wide shedder that reserves capacity for critical traffic, and a sidecar-local in-flight shedder. Thin Go and Python HTTP clients call a per-host Go sidecar, which forwards checks over gRPC to a central Go core that keeps shared state in Redis through atomic Lua scripts. Every tier has a shadow mode. One author, 22 releases through v2.13.1 (2026-09-02).
Tech: Go, Python, gRPC, Redis, Lua, Prometheus, OpenTelemetry, Helm, Docker Compose, Sigstore

### Trelix — Offline code intelligence engine — Tree-sitter indexing, hybrid BM25+vector+grep search
Group: Code Intelligence & Engines · 1556 commits
Repo: https://github.com/sairam0424/trelix
Zero-infra code intelligence — Tree-sitter AST indexing + up to 7-leg RRF retrieval (BM25 + vector + grep + SPLADE + multi-granularity; the graph-BFS leg is opt-in and measured harmful) + knowledge graph (Louvain community detection, Pyvis). MCP server. 4,435 unit test functions on main at v3.4.3. Works offline with no API key.
Tech: Python, tree-sitter, sqlite-vec, SQLite, BM25, FastAPI, LiteLLM, networkx, Qdrant, LanceDB, PyInstaller, Pydantic

### Tombstone — Production intelligence layer for 5,000+ feature flags.
Group: Agent Frameworks & Infrastructure · 1034 commits
Repo: https://github.com/sairam0424/Tombstone
Self-hosted feature flag intelligence with blast-radius gating, circuit-breaker auto-rollback, causal incident correlation ("What Changed?"), and Knight Capital–style tombstoning. Go + Python + TypeScript polyglot.
Tech: Go, Python, TypeScript, React, PostgreSQL, Redis, Kafka, pgvector, Docker, Kubernetes

### MindForge — Agentic-intelligence framework for Claude Code.
Group: Agent Frameworks & Infrastructure · 1822 commits
Repo: https://github.com/sairam0424/MindForge
Command-driven multi-agent orchestration for Claude Code — 221 commands, 164 subagents (152 adapted from VoltAgent, MIT), 354 skills, and cost-aware model routing. Ships as npx mindforge-cc.
Tech: Node.js, TypeScript, MCP, sql.js, SSE

### tracehub-mcp — Read-only MCP server for querying LLM traces across eight OpenTelemetry backends
Group: Platform, Guardrails & Observability · 322 commits
Repo: https://github.com/mcpsmiths/tracehub-mcp
Fork of traceloop/opentelemetry-mcp-server grown into an independently released MCP server for debugging LLM traces: 8 OpenTelemetry backends (Jaeger, Tempo and Traceloop inherited; Datadog, Sentry, AWS X-Ray, New Relic and Honeycomb added) behind one BaseBackend interface, 19 read-only tools (11 inherited) including deterministic root-cause triage and best-effort cross-backend correlation, gen_ai.* attribute parsing with litellm-backed cost attribution, a TTL cache with request coalescing, and SSRF, redirect and response-size guards. At v0.13.0 it is on PyPI, the MCP registry and GHCR; 303 of its 322 commits post-date the fork.
Tech: Python, FastMCP, MCP, OpenTelemetry, Pydantic, httpx, boto3, Hypothesis, Docker, uv

### Anvilry — Portfolio with four switchable views and a read-only MCP server over one MDX source
Group: Tooling & Lab · 865 commits
Repo: https://github.com/sairam0424/anvilry
A personal engineering portfolio with four switchable experiences over one Velite content source: a static Classic site, an AI concierge chat grounded on the résumé corpus through AWS Bedrock with a model fallback chain and a deployment-tagged FAQ cache, a WebGL build graph, and a keyboard-native developer terminal. Agents read the same content through a read-only MCP server, llms.txt and resume.json. A fingerprinted citation gate, a hash-chained claims ledger and a first-load bundle budget check the docs and numbers in CI; none is a required merge check.
Tech: Next.js, React, TypeScript, Tailwind CSS, Velite, AWS Bedrock, Upstash Redis, React Three Fiber, MCP, Vitest

### Agent-Forge — Self-improving AI agent infrastructure.
Group: Agent Frameworks & Infrastructure · 209 commits
Repo: private repository
A Karpathy-style propose/eval/score/commit-or-revert loop over a git-versioned agent spec, with an LLM-judge eval harness. Framework-agnostic.
Tech: Python, FastAPI, pgvector, Celery, MCP, Docker

### Graph-Forge — AI-native distributed code-intelligence platform.
Group: Code Intelligence & Engines · 574 commits
Repo: private repository
Models code as a Neo4j knowledge graph + Chroma embeddings over 19 polyglot microservices, powering RAG over massive codebases with full observability.
Tech: Go, Python, gRPC, Neo4j, Chroma, Kafka, Next.js

### Inkforge — Notes, topics, or code → published technical articles.
Group: Tooling & Lab · 124 commits
Repo: https://github.com/sairam0424/Inkforge
STORM two-stage pipeline (outline → draft → polish) with BM25 in-memory RAG, streaming web UI, and one-command publishing to Dev.to (API), plus Medium and Hashnode publishers that script your own logged-in Chrome (not live-verified); Substack and LinkedIn stay manual.
Tech: TypeScript, Node.js, Next.js, Turborepo, Bedrock, BM25, Commander, Vitest

### ContextOS — Intelligence layer for autonomous AI agents.
Group: Agent Frameworks & Infrastructure · 233 commits
Repo: https://github.com/sairam0424/ContextOS
SQLite/vector indexing, multi-agent orchestration, and resilience — shipped as core/CLI/MCP npm packages plus a 3D spatial dashboard.
Tech: TypeScript, SQLite, Vector Search, MCP, React, Three.js

### ag-bash — AI-native bash interpreter, in TypeScript.
Group: Code Intelligence & Engines · 512 commits
Repo: https://github.com/sairam0424/ag-bash
A sandboxed bash interpreter for AI agents — Tree-sitter WASM parser, in-process WASM runtimes (CPython, QuickJS, SQLite3), exposed over MCP.
Tech: TypeScript, WebAssembly, MCP, Tree-sitter, pnpm

### CommandVault — Universal AI command manager.
Group: Tooling & Lab · 194 commits
Repo: https://github.com/sairam0424/CommandVault
Indexes 350+ slash commands, skills, agents, and plugins across Claude Code, Cursor, Copilot, Windsurf, and Aider into one searchable vault.
Tech: TypeScript, VS Code Extension, CLI, SQLite

### Daily Dose — AI-curated daily tech digest: four sources, one batched Bedrock call, per-run cost ledger
Group: Tooling & Lab · 179 commits
Repo: https://github.com/sairam0424/daily-dose
A cron-driven Astro static site that pulls Hacker News, arXiv, GitHub and Dev.to through free keyless APIs, scores all twenty daily items in one forced-tool-use AWS Bedrock call (Sonnet 5 leading a four-model fallback chain), and validates the output against one Zod schema shared by the pipeline and the Astro content collection. GitHub Actions commits the result back to `main` each night. Every run's token usage and dollar cost goes into a committed JSONL ledger (52 runs, $3.10 total at 2026-10-20) with a 5x trailing-average anomaly flag that files a GitHub Issue.
Tech: TypeScript, Astro, Node.js, Bedrock, Zod, Vitest, GitHub Actions, Vercel, axe-core

### nh-deck — Local-first CLI that turns one Markdown file into a presentable slide deck
Group: Tooling & Lab · 58 commits
Repo: https://github.com/sairam0424/nh-deck
One Markdown file in, a slide deck out: served on 127.0.0.1 with live reload, presented in the browser, or exported to PDF, per-slide PNGs or a static index.html, all from a single generateHtml() pipeline. Slides split at marked's hr tokens, KaTeX fonts ship as base64, the Google Fonts @import is stripped from every Mermaid SVG, and PDF export drives a Chrome-family browser the user already has through puppeteer-core, so there is no bundled browser, no CDN and no telemetry. 13 ADRs and 563 Vitest tests back it. The repo is tagged v1.10.0; npm still serves 1.7.0.
Tech: TypeScript, Node.js, Commander, marked, puppeteer-core, KaTeX, Mermaid, highlight.js, Vitest, GitHub Actions

## Skills
Languages: Python, Go, TypeScript
GenAI: LLM Agent Orchestration, Multi-Agent Systems, RAG, Prompt Engineering, crewAI, Vector DBs & Embeddings
Backend & Distributed: FastAPI, gRPC, REST, GraphQL, Microservices, Event-Driven
Data & Messaging: PostgreSQL, MongoDB, Redis, ChromaDB, Apache Kafka, Redis Streams, SSE, WebSockets
Cloud & Ops: AWS, Docker, Kubernetes, CI/CD, OpenTelemetry, Prometheus, Grafana
Frontend: React, Angular, Next.js, Tailwind CSS

## Achievements
Google Code Jam 2023 — AIR 420 · 3,687 / 85,000+
Meta Hacker Cup 2022 — 4,048 / 70,000+
Flipkart GRiD 2022 — Top tier · 1,325 / 40,000+
Institute Rank 1 — GeeksforGeeks & InterviewBit
Mentored 250–300 students — Data Structures & Algorithms

## Personal (beyond the résumé)
Hobbies: Competitive programming; Building & customizing mechanical keyboards; Reading — tech, sci-fi, and non-fiction; Food & travel
Fun facts: Solved 1000+ data-structures & algorithms problems; Mentored 250–300 students on DSA
Currently learning: Rust & Go for systems-level work; Distributed consensus protocols; System design & architecture at scale; Deeper LLM / agent internals
Ask me about: Multi-agent orchestration tradeoffs; Scaling an event-driven backend to 3K daily users; Redis Streams vs Kafka for streaming; Mentoring & teaching DSA
Uses: Editor & terminal — VS Code, Cursor, Warp, iTerm2 + zsh + tmux; Hardware — MacBook Pro (Apple Silicon), A mechanical keyboard; Infra & ops — Docker, Kubernetes, Argo CD, AWS, Azure; Daily tools — Git & GitHub, Postman, Notion, Obsidian
Right now: Pensieve & AAVA Code work at Ascendion; Architecting solutions across work and open source; Building open-source AI infrastructure; Deepening systems & distributed-systems knowledge — open to new roles

## Writing
### I Built trelix Because I Was Tired of Grepping My Way Through Codebases
How I built trelix — a zero-infra code intelligence engine that indexes any repo with Tree-sitter, runs 7-leg RRF retrieval, and answers questions about your codebase offline. From four hours of archaeology to a query that returns in 200ms.

### We Shipped Tombstone v1.0. Then We Found Nine Bugs That Would Have Paged Us at 2am.
What happened when we pressure-tested Tombstone v1.0 against production reality: a kill switch silently returning 400, approval endpoints wired to routes never registered, a scheduler that could double-fire. Nine bugs. Nine fixes. Here's the post-mortem.

### How I Built Tombstone: A Self-Hosted Feature Flag Intelligence Platform to Prevent the Next Knight Capital
A first-person engineering account of building Tombstone — an 8-service, self-hosted feature flag platform with tombstoned keys, causal incident correlation, and circuit-breaker auto-rollback — born from a 2am on-call incident and the realization that Jira plus Notion is not an audit trail.

### How I Built Inkforge: Designing an AI-Powered Article System with STORM Pipeline, BM25 RAG, and AWS Bedrock
A first-person engineering deep-dive into building Inkforge — an AI article generation system that combines the STORM research pipeline, BM25-based retrieval-augmented generation, and AWS Bedrock to produce structured, grounded long-form content at scale.

### Feature Flags at Scale: Designing a Distributed Control System for Production Behavior
At scale, feature flags stop being booleans and become a distributed control plane — here's the architecture, failure modes, and operational patterns that make them work at millions of evaluations per second.

### How I Traced One Browser Request from Keystroke to Rendered Page
A packet-level walkthrough of what actually happens when you type www.google.com — DNS, TCP, TLS, HTTP, and rendering — with the latency costs and design trade-offs at each layer.

### How DNS Actually Works: Resolution Hierarchy, Caching, and Production Failure Modes
A systems-level walkthrough of DNS resolution — from browser cache to authoritative nameserver — covering record types, anycast infrastructure, GeoDNS, and the operational edge cases that bite engineers in production.