Senior Applied AI / LLM EngineerFounding Team (Commerce AI)
Join Briskk.ai and help build a WhatsApp-native commerce revenue engine for high-ticket omnichannel retail.
Applied AI / LLM
Build production-grade LLM workflows for commerce
Backend & Data
Postgres, pgvector, Redis, Elixir/Phoenix
Founding Team
Help define product, stack, and culture
About Briskk.ai – The World You're Entering
Most serious purchase decisions in India don't happen on a website. They happen through conversations – on WhatsApp or in-store with staff.
Yet high-ticket, omnichannel brands still run on:
- Leaky funnels between WhatsApp, store visits, and websites
- Staff trying to remember every walk-in and follow-up
- CRMs that don't understand SKUs, carts, or real conversations
- Stateless, rule-based chatbots that can blast offers but can't close
Briskk.ai is building a WhatsApp-native commerce revenue engine for omnichannel, high-ticket retail (starting with furniture on Shopify).
On top of Shopify + in-store data, we run two agents on a shared brain:
- 🛒Customer Agent – helps end customers search, compare, and ask about products directly on WhatsApp.
- 🧑💼Staff / Lead Agent – lets store staff capture, qualify, and update leads (with full product context) from the same thread.
One shared state, two agents, zero excuses for lost leads.
We've already:
- Built a multi-tenant Elixir/Phoenix + Postgres (JSONB, FTS, pgvector) + Redis platform
- Cloned the Shopify data model on our side
- Wired it into a real WhatsApp agent that runs end-to-end demos on live furniture catalogs
- Live with paying furniture brands, handling thousands of WhatsApp conversations and generating revenue across customer + staff interactions
Now we're pushing from working product → undeniable revenue impact → PMF and seed round.
You'll join exactly at this inflection point.
The Quest – Your Role
This is not a pure ML research role and not a simple backend CRUD role. You'll sit at the intersection of:
- Applied LLM systems – tools, workflows, evals, guardrails
- Backend & data infra – APIs, Postgres/pgvector, Redis, multi-tenant state
- Commerce domain – Shopify, carts, orders, leads, staff workflows
Your mission:
Turn our current agents into a reliable, measurable revenue engine that D2C brands can't imagine turning off.
You'll work directly with the founder (ex-Coinbase, ex-Walmart, ex-Urban Ladder) and:
- Own the LLM workflows that power customer + staff conversations
- Own the search & ranking stack combining FTS + pgvector
- Own the reliability, latency, and observability of the WhatsApp agent in production
This is a founding team seat – you are not joining as "employee #27", you're helping define the product, the stack, and the culture.
Quests You'll Undertake
1. Design and build agentic workflows (Two Agents, One Brain)
- • Architect and own the core logic for the Customer Agent and Staff / Lead Agent
- • Intent detection and routing between product mode and lead/CRM mode
- • Multi-step tool usage (product search, inventory, pricing, lead update, etc.)
- • Conversation state and memory (Redis, JSON state/frames, tenant-aware context)
- • Use OpenAI (and similar) APIs to design robust prompts and structured outputs / tools
- • Minimize hallucinations and brittle flows
- • Control latency and token usage like a performance engineer
- • Build evaluation harnesses for critical intents with automated scoring
2. Own the hybrid search & ranking engine
- • Implement and tune Postgres text search (tsvector, ts_rank_cd, trigram similarity)
- • Extend and optimize pgvector for semantic and "similar items" search
- • Hybrid pipeline: keyword filter → vector rerank
- • Latency-conscious design that stays under a strict p95 budget
- • Expose clean, composable APIs for natural language discovery
3. Forge and harden backend APIs for the agent
- • Extend our Elixir/Phoenix + Postgres + Redis stack
- • Design and implement APIs for product search, lead management, and staff workflows
- • Keep the system secure, multi-tenant, and predictable under real-world traffic
4. Make the AI truly production-grade
- • Instrument everything with metrics and logs (p50/p95 latency, error rates)
- • Create dashboards and runbooks for production systems
- • Track metrics that actually matter: leads created, qualified, closed, conversion rates
Target: ≤2.5s p95 response time end-to-end on WhatsApp across catalogs of 5k–50k SKUs per tenant
What success looks like in your first 90 days
Days
You understand our current agent flows, ship fixes to the top 2–3 reliability/latency issues, and help shape evaluation harness v1.
Days
You own a slice of the stack (e.g., hybrid search or lead APIs) and ship a visible v2 with better p95 and debuggability.
Days
You're the clear owner for at least one core area (search, lead flows, or eval/observability), and we can see your work reflected in production metrics.
Requirements & Qualifications
You don't have to tick every box, but this describes you reasonably well:
You're likely a strong fit if you've:
- ✓ Shipped at least one LLM-powered workflow with tools / structured outputs into production
- ✓ Owned Postgres performance on real traffic (indexes, EXPLAIN, query tuning)
- ✓ Worked on systems where p95 latency and reliability actually mattered
Required Experience:
- • 4–9 years in backend / platform / applied ML/AI roles
- • Strong in at least one backend language (Elixir/Phoenix ideal, or Python/FastAPI, Node.js, Go)
- • Deep comfort with Postgres (query design, indexes, JSONB, full-text search)
- • Hands-on experience with Redis (key spaces, TTL strategies, caching)
- • Real experience building LLM-powered applications with OpenAI/Claude
- • Think in terms of latency and reliability (p50/p95, retries, timeouts)
- • Product-minded: care about what gets brands more leads and sales
Nice-to-Haves (Bonus Points):
- • Experience with Shopify Admin API or other commerce platforms
- • Prior work in conversational commerce or CRM/lead management tools
- • Experience with pgvector or other vector databases
- • Prior early-stage startup or founding-team experience
Your Arsenal (Stack)
Backend
- • Elixir/Phoenix, Oban
- • Python/FastAPI for agent services
Database
- • Postgres (JSONB, FTS, pgvector)
- • Redis
AI/LLM
- • OpenAI (GPT-class models)
- • Structured outputs / tools
- • Eval harnesses
Integrations
- • Shopify Admin API
- • WhatsApp Business API/BSPs
Infrastructure
- • Docker, AWS (EC2/RDS/Redis), GitHub Actions CI/CD
How We Work
- • Tiny team, high-bandwidth collaboration with the founder
- • Tight feedback loop with paying customers (live in furniture retail)
- • Preference for small, frequent releases over big-bang refactors
- • You'll be involved in architecture, design discussions, and customer calls
We care about:
- • Clarity: simple designs, simple APIs, clear ownership
- • Reliability: things should work, not just look good in a demo
- • Speed: deciding and shipping in days, not quarters
Compensation & Upside
We're pre-seed and pre-PMF, but with paying customers and thousands of conversations.
We'll be upfront:
We cannot match Big Tech or late-stage startup salaries right now. Cash is constrained until we scale revenue beyond furniture and close our first round.
What we can offer:
- Cash: Lower than Big Tech / late-stage startup to begin with, with room to step up as revenue lands. Options include:
- • A modest fixed base, plus project / goal-based incentives, or
- • A contract-to-hire path for 2–3 months, then stepping up as revenue scales
- Equity: Meaningful upside as part of the founding team (roughly 1–2.5% ESOP range for this seat, with milestone-based top-ups possible as we scale)
- • Standard 4-year vesting with 1-year cliff
- Upside path: As we hit PMF and raise a round, this role can grow into:
- • Founding Engineer / Head of Engineering, or
- • Lead for Applied AI / Agent Systems
We'll be transparent about numbers, runway, and milestones from day 1, so you can make an informed decision about the risk/reward curve you're signing up for.
How to Join the Quest
If this sounds like the kind of risk/reward curve you want to ride:
Send us:
- • Your GitHub/portfolio, or 1–2 repos you're proud of
- • Any LLM/AI-powered project you've worked on (internal or side project)
- • 2–3 short paragraphs on:
- – A system you've designed end-to-end, and
- – Either: (a) how you'd design the retrieval pipeline for: "L-shape sofa under 45k, fabric, 84 inches, in stock in HSR store", or (b) how you'd measure lead leakage and WhatsApp-driven uplift for a furniture brand
We'll do a quick intro call, a deep-dive tech discussion, and then a short, paid trial project so you can see the real system before committing.
Total timeline: Usually ~2–3 weeks from first call to offer, including a short paid trial project.
ChannelBlend welcomes individuals of all backgrounds. We're committed to crafting a place where builders, debuggers, and dreamers can do the best work of their careers – and have a real shot at changing how retail works in India and beyond.
Ready to Join the Quest?
Help us build the commerce AI brain that will power the next generation of retail in India.