Chat and RAG agents
Ground answers in your documents. Tool calls into CRM and internal APIs. Guardrails, escalation to humans, and evaluation so the agent stays reliable in production.
AI agents · Voice · RAG · Full-stack delivery
Build agents that ship: policy decides first, a guard checks every sentence, and evals prove it.
I design and deliver production AI agents: chat, voice, RAG, CRM tool-calling, and multi-agent workflows. Proof includes Darwin AI, LiveKit Agents, Pipecat, Genkit, and Swarms. Full-stack Angular and React when the product needs a UI.
AIagents primary offer
9OSS agent fixes
4+years shipping
telemetry · measured, not claimed
Deterministic eval agent-receptionist
44/44
Hotel 23/23, dental 21/21.
Live Groq eval 2026-10-24
43/44
One miss: an eval regex too narrow, reply was right.
Live 20-question test 2026-10-24
19/20
Same cause: too-narrow regex, correct reply.
First token live check
~0.8s
~840 ms on the public URL. Real token streaming, not one buffered reply.
Cost / question live check
~$0.0002
563 prompt + 251 completion tokens. A single check, not a benchmark.
Policy-routed replies pre-router
0tokens
Emergencies, hours and handoff never reach the model.
Throughput bleedstop
6,291decisions/s
p99 32 ms on 3 instances.
Rolling restart bleedstop · k8s
0/ 287,403 failed
Requests during a Kubernetes rolling restart. 0% false positives on benign eval scenarios.
01 · services
AI agent services
Primary offer. I scope chat, voice, RAG, and CRM-connected agents for production, then ship the product UI around them when needed.
Voice agents
Phone and in-app voice with interruption handling, tool use, and clean call teardown. Hands-on with LiveKit Agents and Pipecat reliability paths.
Multi-agent workflows
Specialized workers for inbound, outbound, support, and collections-style flows. Experience across Genkit, Swarms, and harness SDKs.
Product UI around the agent
Angular and React interfaces, Node and .NET APIs, Docker and AWS, so the agent is part of a shippable product, not a demo notebook.
02 · projects
AI agent projects
Detailed proof behind the pitch.
AI agent · engineering case study
Agent Receptionist
Tool-calling LLM agent (Groq, gpt-oss-120b) for appointment-based businesses. RAG with cited sources, policy pre-routing for emergencies, hours and handoff at zero tokens, a per-sentence guard, a deterministic booking fallback, and streaming SSE replies. Measured: deterministic eval 44/44, live Groq eval 43/44 (one eval-regex miss), live 20-question test 19/20, about $0.0002 per question in one live check. Demos use fictional businesses; source is private, available on request.
- before modelPolicy pre-router. Rules decide emergencies, hours and handoff.
- after modelPer-sentence guard. Entity allowlist, hours check, required-fact check.
- securitySecrets and raw provider errors stay server-side.
Watch the demo. A 73 second walkthrough with voiceover: the case study, the live demo (tool calling, RAG with sources, latency and cost per turn) and the same engine on a different fictional business.
01
BleedStopLLM security · infrastructure
Cost abuse control plane for LLM, SMS and payment APIs (OWASP LLM10). Apps ask before they spend; deterministic multi tenant rules, Redis atomic budgets, Postgres durable mode, agent guard, Node and Python SDKs, Helm on Kubernetes. Measured: 6,291 decisions/s at p99 32 ms on 3 instances, 0 failed of 287,403 requests during a Kubernetes rolling restart, 0% false positives on benign eval scenarios.
02
Jivi AI medical agentsHealthcare · multi-agent
Multi-agent healthcare system for doctors, patients, and providers. Clinical agents for triage, risk, treatment, medication safety, radiology, ECG, and labs. Wellness agents for nutrition, stress, and mental health. Doctor Co-Pilot for summaries, auto-scribe, diagnosis support, and medical coding.
03
Darwin AIProduction product
Angular UI for AI employees that run customer conversations on WhatsApp, Instagram, and phone. Specialized roles for inbound, outbound, CX, collections, and post-sales, with CRM updates and human escalation.
04
LiveKit AgentsVoice · contribution
Bounded EndCallTool wait for tool-reply audio. Skipped the drain when that reply times out so voice sessions do not hang on teardown.
05
PipecatRealtime · contribution
Skipped Responses drain on early cancel. Kept function-call timeouts armed across intermediate results. Added sequencing on turn-stopped messages.
06
Genkit · Swarms · harnessFrameworks
Genkit: decode base64 OpenAI embeddings for retrieval. Swarms: optional swarm description for multi-agent prompts. Harness SDK: JSON-safe AgentResult with bytes.
07
AI code reviewerNode
Express and Apollo GraphQL service with MySQL and SQL Server, plus scripts that check code quality for AI-assisted review workflows.
08
Darwin StudioGenAI product
IT Pixelz. Angular, Node, MySQL. Prompted generation with ChatGPT and Stability AI, plus upscaling, model switching, and image-to-motion.
03 · product work
More product work
Angular and full-stack delivery alongside agent work.
09
ERA Citizen ScienceAngular · live
Angular work on Malta Environment and Resources Authority citizen-science platform.
10
Doctor appointmentsMERN
React, Node, Express, MongoDB, Tailwind. Book by specialization. Admin adds doctors.
11
IT Staffing SolutionsStaffing
About 1.5 years. Angular and Bootstrap, then WordPress, Strapi, and API modules.
12
Angular and React appsFrontend
Ecommerce, book store, LMS, portfolio, and more under GitHub. Full list in the Angular and React section below.
04 · open source
Open source I worked on
Contributions. I didn’t start these — I shipped the fix named here.
LiveKit Agents
Bounded EndCallTool wait; skipped drain on tool-reply timeout. Fork
Pipecat
Skip Responses drain on early cancel; keep function-call timeouts armed; sequence turn-stopped. Fork
Genkit · Swarms · harness SDK
Embeddings decode · optional swarm description · JSON-safe AgentResult with bytes.
TMail · Drift · Palinode · Celluloid
Reply-to fallback + tests · FAQ · CLI exit codes · --mpv-pause flags.
05 · interfaces
Angular & React
More interface work on my GitHub.
06 · experience & skills
About four years of professional delivery
Based in Islamabad, Pakistan. Remote-friendly with distributed teams. Timezone PKT, UTC+5.
- Jul 2024 → presentAngular developer · Onlinx SolutionsCurrent role.
- Agent workJivi AI medical agentsMulti-agent clinical, wellness, and doctor co-pilot systems. jivi.ai/ai-agents
- Mar 2023 → Jun 2024Angular · IT PixelzCRM dashboards, appointment scheduling, and reporting, plus Darwin Studio GenAI work.
- Jan 2022 → Feb 2023Junior Angular developer · Zepto Systems
- 2017 – 2021BS Computer Science · Allama Iqbal Open UniversityIslamabad. Additional coursework: MEAN bootcamp with DevNation, Frontend Masters courses on Git and JavaScript design patterns.
AI and agentsLLMs, RAG, chat agents, voice agents, multi-agent systems, LiveKit Agents, Pipecat, Genkit, Swarms, harness SDKs
FrontendAngular, React, TypeScript, RxJS
BackendNode, Express, .NET, FastAPI, REST, GraphQL
DataSQL, MongoDB, MySQL, SQL Server
DeliveryDocker, AWS, Git, CI/CD
AlsoNode.js, RabbitMQ, Docker / Nginx, vector DBs, LangChain, Ollama
Need an agent that holds up in production?
I scope price and timeline after a short brief. Source for Agent Receptionist is private, available on request or in interviews. Write to me directly.
- usmamalik346@gmail.com
- phone
- +92 306 9607679
- in/osamamalik346
- github
- usmamalik345