Parbhat Kapila

Work

• Sentinel

  

  01SentinelRevenue Intelligence · Permission Layer

  A revenue-intelligence backend whose first question is what it's even allowed to read. Email, calendar, and chat are ingested only when a participant matches a contact synced from your CRM - read-only, fail-closed, and PII-blind.

  • CRM matchrequired on every inbound item
  • Fail-closedon any lookup error
  • Read-onlynever writes to the CRM
  • 2,200+tests
• Cutline

  

  02CutlineAI Video Pipeline · 11-12 Stage Director

  One sentence in, one finished MP4 out - directed by a pipeline of up to 12 stages per slideshow, not a template engine. It infers audience, goal, tone, and arc; plans 1-3 shots per story beat, with more beats for longer videos; sources or generates imagery; composes the render. No templates, no creative knobs.

  • 11-12 stagesper fresh slideshow render
  • 1st saleSep 2026, inbound
  • SHA-256avatar cache in Redis
  • Event-keyedpayment webhook dedup
• RepoDoc

  

  03RepoDocCodebase RAG · Indexing Infrastructure

  Ask an unfamiliar repo questions and get grounded, cited answers. Retrieval runs over what each file means - LLM summaries embedded, not raw code - indexing is a durable Postgres lease queue that survives serverless limits, and each answer's estimated model cost is recorded per project.

  • 5-minjob leases
  • 45swork budget under a 60s cap
  • 5 retriesbefore a job dies
  • 1 tarballper pass, pinned to a commit
• VectorMail

  

  04VectorMailEmail Agent · Rules Decide, LLM Drafts

  Reads your inbox, drafts follow-ups, and sends them. Deterministic rules pick the threads: a reminder is due, or the last message came from an external sender and you haven't replied. Rules also decide whether each follow-up sends, waits for your approval, or is dropped; the LLM only writes the draft. Inbox and embeddings live in one Postgres, with no separate vector store.

  • One DBinbox and vectors
  • Dry-runby default
  • 2 gatesbefore an AI reply sends
  • Tenancyin the same WHERE clause
• Visura

  

  05VisuraDocument Versions · Hashes Decide, LLM Explains

  Shows what changed between two versions of a document, in your browser, with no upload and no model. Passage boundaries come from each sentence's own hash, so an edit near the top doesn't shift the rest, and a re-upload re-summarizes only what changed. The LLM only explains changes the hashes found, and drops any explanation it can't quote from the document.

  • 19 of 20reused after a one-line edit
  • 12 re-exportsall read as zero changes
  • No uploadand no model to compare
  • Verbatim quoteor the explanation is dropped

Walkthroughs

Walkthroughs I recorded on Loom for Sentinel, Cutline, RepoDoc and VectorMail, each 11 to 14 minutes long. Open any one to watch.

• 11:1601SentinelA revenue-intelligence backend whose first question is what it's even allowed to read.Watch walkthrough
• 11:2202CutlineOne sentence in, one finished MP4 out - directed by a pipeline of up to 12 stages per slideshow, not a template engine.Watch walkthrough
• 13:4003RepoDocAsk an unfamiliar repo questions and get grounded, cited answers.Watch walkthrough
• 13:4104VectorMailReads your inbox, drafts follow-ups, and sends them.Watch walkthrough

SentinelCutlineRepoDocVectorMail

Sentinel · Loom walkthrough · 11:16

1×1.25×1.5×2×

Writing

• 01Sentinel: CRM is the permission layerRevenue-intelligence runs on the most sensitive data a company holds. So the first question isn't "what model scores the deal" - it's "what are we even allowed to read?"Read the case study →
• 02CUTLINE: a director, not a template engineOne sentence in, one finished MP4 out - directed by a pipeline of up to 12 stages per slideshow, not a template engine. The pipeline commits the editorial decisions before it touches a frame.Read the case study →
• 03RepoDoc: codebase RAG built as infrastructureRetrieval runs over what each file means, indexing is a durable Postgres lease queue, and each answer's estimated model cost is recorded per project.Read the case study →
• 04VectorMail: rules decide, the LLM only draftsReads the inbox, drafts follow-ups, and sends them. Deterministic rules decide, the LLM only drafts.Read the case study →

Experience

Backend Engineer

FunnelStory · Remote · Nov 2026 - Present

Backend engineer on the Go platform behind FunnelStory's AI outcome agents for post-sales revenue teams.

Working on the data layer: how customer data gets pulled in from warehouses and SaaS tools and turned into the account and activity data the agents act on.

• Go
• PostgreSQL